JOURNAL OF NEUROTRAUMA 30:1781–1794 (November 1, 2013)

ª Mary Ann Liebert, Inc. Review

DOI: 10.1089/neu.2013.2932

The Effects of the Timing of Spinal Surgery

after Traumatic Spinal Cord Injury:
A Systematic Review and Meta-Analysis

Joost J. van Middendorp,1–3 Allard J.F. Hosman,4 and Suhail A.R. Doi 3

Abstract
The debate over the effects of the timing of surgical spinal decompression after traumatic spinal cord injury (tSCI) has
remained unresolved for over a century. The aim of the current study was to perform a systematic review and quality-
adjusted meta-analysis of studies evaluating the effects of the timing of spinal surgery after tSCI. Studies were searched
for through the MEDLINE database (1966 to August 2012) and a 15-item, tailored scoring system was used for assessing
the included studies’ susceptibility to bias. Random effects and quality effects meta-analyses were performed. Models
were tested for robustness using one way and criterion-based sensitivity analysis and funnel plots. Results are presented as
weighted mean differences (WMDs) and odds ratios (ORs) with 95% confidence intervals (CIs). A total of 18 studies were
analyzed. Heterogeneity was evident among the studies included. Quality effects models showed that – when compared
with ‘‘late’’ surgery – ‘‘early’’ spinal surgery was significantly associated with a higher total motor score improvement
(WMD: 5.94 points, 95% CI:0.74,11.15) in seven studies, neurological improvement rate (OR: 2.23, 95% CI:1.35,3.67) in
six studies, and shorter length of hospital stay (WMD: - 9.98 days, 95% CI: - 13.10, - 6.85) in six studies. However, one
way and criterion-based sensitivity analyses demonstrated a profound lack of robustness among pooled estimates. Funnel
plots showed significant proof of publication bias. In conclusion, despite the fact that ‘‘early’’ spinal surgery was
significantly associated with improved neurological and length of stay outcomes, the evidence supporting ‘‘early’’ spinal
surgery after tSCI lacks robustness as a result of different sources of heterogeneity within and between original studies.
Where the conduct of a surgical, randomized controlled trial seems to be an unfeasible undertaking in acute tSCI, meth-
odological safeguards require the utmost attention in future cohort studies. (Prospero registration number: PROSPERO
CRD42012003182. See also http://www.crd.york.ac.uk/NIHR_PROSPERO/)

Key words: meta-analysis; SCI; systematic review; timing of surgery

Introduction edema, increased excitatory amino acids, and lipid peroxidation.2

Pre-clinical data support the theory that persistent compression of

M ore than a century ago, Burrell reported two key

issues related to the surgical management of traumatic spinal
cord injury (tSCI): the timing of surgery and the severity of the
injury.1 Nowadays, the topic of the timing of surgical spinal de-
compression after tSCI continues to spark vigorous debates among
specialists.
Advocates for early spinal decompressions refer to the concept
of primary versus secondary mechanisms of injury.2 The primary
mechanism comprises the initial cord lesion that results from
physical trauma to the tissue caused by a displacement of sur-
rounding spinal structures. The primary mechanism in turn initiates
a cascade of secondary injury mechanisms including ischemia,
the spinal cord represents a cause of secondary injury and, there-
fore, may be potentially reversible.3,4 Despite positive effects of
acute spinal decompressions reported in these standardized pre-
clinical studies, no neurological benefits have consistently been
reported in the human, clinical setting.
A number of systematic reviews examining the timing of spinal
surgery after tSCI have been published over the last decade.5–10 A
profound limitation of all of these reviews is that non-comparative
case series were also included. Furthermore, the meta-analysis
published by La Rosa and colleagues10 was severely limited by
applying a single, nonspecific pooled outcome measure, namely the
‘‘neurological improvement rate.’’ Finally, the impact of original

1
Stoke Mandeville Spinal Foundation, National Spinal Injuries Centre, Stoke Mandeville Hospital, Aylesbury, United Kingdom.
2
Harris Manchester College, University of Oxford, Oxford, United Kingdom.
3
Clinical Epidemiology Unit, School of Population Health, University of Queensland, Herston Road, Brisbane, Australia.
4
Spine Unit, Department of Orthopaedics, Radboud University Nijmegen Medical Center, Nijmegen, The Netherlands.

1781
1782
studies’ variable susceptibility to bias on pooled outcome estimates
has not been considered in previous reviews.
The aim of the current study was to perform a systematic review
and meta-analysis of the effects of the timing of spinal surgery after
tSCI that differs from previous reviews in two ways. First, we critically
appraised the reporting of key methodological safeguards reducing the
susceptibility to bias in studies comparing outcomes in patients who
underwent either ‘‘early’’ or ‘‘late’’ spinal surgery. Second, based on
the methodological quality score of each individual study, we subse-
quently performed a quality effects (QE) meta-analysis.
Methods
Data sources
In order to identify relevant articles on the effects of the timing
of spinal surgery after tSCI, we conducted a computerized search
using the MEDLINE (1966 to August 2012) database. The search
terms used in the PubMed interface are presented in Supplementary
Appendix 1 (see online supplementary material at http://www
.liebertonline.com). To ensure that no relevant studies were missed
by the computerized search, we also performed a manual cross-
reference search of the citations of each included article to obtain
further relevant studies.11
FIG. 1. Literature search and selection process. (*References that fit into more than one exclusion criterion were categorized in a
single category at the abstractor’s discretion, {Fourteen of the 18 studies [18%] included more than one outcome measure suitable for
meta-analysis.)
VAN MIDDENDORP ET AL.
Eligibility criteria
The initial screening for eligibility included the following two
criteria. The study had to consist of clearly defined cohorts of
subjects who had undergone spinal surgery after tSCI. In addition,
comparative outcomes of ‘‘early’’ and ‘‘late,’’ or delayed, spinal
surgery had to be reported. Both randomized and nonrandomized
comparative studies were eligible for inclusion. All neurological
levels and all severities of tSCI were included. All approaches for
spinal surgery were eligible for inclusion, including anterior, pos-
terior, and circumferential open approaches, as well as closed
manual reductions. Studies combining patients with non-tSCIs with
patients with tSCIs were excluded. Studies that included pa-
tients <14 years of age were also excluded. Articles published in
languages other than English and those without an abstract were
excluded. Titles and abstracts of retrieved articles were screened,
and potentially relevant full-text articles and reports were retrieved
and evaluated individually and independently by two reviewers
( J.J.v.M. and A.J.F.H.), who were trained and experienced in
performing systematic literature searches. In each phase, and in all
cases, disagreement concerning inclusion of articles was resolved
by discussion and consensus agreement. Details of the litera-
ture search and selection process are outlined in the flow chart in
Figure 1.
TIMING OF SPINAL SURGERY AFTER tSCI 1783

Data abstraction and quality assessment
Critical appraisal of included articles was performed to examine
studies’ susceptibility to bias. A standard data extraction form
summarizing the study design, study population, and relevant raw
data was completed for each article. Because a standardized as-
sessment for the methodological quality of both surgical RCTs and
non-RCTs does not currently exist, a tailored bias assessment
scoring instrument was devised based on reworded methodological
items covered by the following three consensus-derived ‘‘check-
lists’’ for reporting randomized and nonrandomized clinical
studies: the Consolidated Standards of Reporting Trials of Non-
pharmacologic Treatment (CONSORT-NPT) checklist,12 the cri-
teria list for methodological quality assessment introduced by the
Cochrane Collaboration Back Review Group for Spinal Dis-
orders,13 and the Strengthening the Reporting of Observational
Studies in Epidemiology (STROBE) statement.14
A 15-item, tailored scoring system was devised for the assess-
ment of comparative surgical studies in spinal trauma (Table 1).
The minimum and maximum quality scores are 0 and 25 points,
respectively. Included items cover key methodological safeguards
that reduce studies’ susceptibility to bias. As assessment of the

Table 1. A Tailored Checklist for Assessing the Susceptibility

to Bias in Therapeutic Comparative Studies for Spinal Cord Injury

Item Question Answer (score)

Selection bias

1 Was the treatment group allocation at random? Yes (1), No/Unclear (0)
2 Was the allocation sequence generated and concealed adequately? Yes (1), No/Unclear (0)
3 Was attrition < 20% and if so, was attrition comparable between both groups? Yes (1), No/Unclear (0)
4 Was the length of follow-up equal between the groups? Yes (1), No/Unclear (0)
For case-control studies: Was the time period between the intervention/exposure and
outcome the same for cases and controls?

Information bias

5 Were data assessed and recorded prospectively in both study groups? Yes (1), No/Unclear (0)
6 Were the primary and secondary outcomes and related hypotheses pre-specified? Yes (1), No/Unclear (0)
7 Were interventions clearly defined and comparable between and/or within the groups? Yes (1), No/Unclear (0)
8 Were the following persons blinded for treatment allocation?
a) Participants Yes (1), No/Unclear (0)
b) Care providers Yes (1), No/Unclear (0)
c) Outcome assessors Yes (1), No/Unclear (0)
9 Were standardized, reproducible definitions used for:
a) Baseline characteristics Yes (1), No/Unclear (0)
b) Study outcomes Yes (1), No/Unclear (0)
c) Adverse events Yes (1), No/Unclear (0)
10 Apart from blinding, were other safeguards used for assuring the reliability of:
a) Baseline characteristics Yes (1), No/Unclear (0)
b) Study outcomes Yes (1), No/Unclear (0)
c) Adverse events Yes (1), No/Unclear (0)

Confounding

11 Were important prognostic factors comparable in both groups or stratified/

matched/adjusted for analysis?
a) Severity of SCI Yes (1), No/Unclear (0)
b) Level of SCI Yes (1), No/Unclear (0)
c) Concomitant traumatic injuries Yes (1), No/Unclear (0)
d) Pre-existing comorbidities Yes (1), No/Unclear (0)
e) Age Yes (1), No/Unclear (0)
12 Did researchers rule out any impact from a concurrent intervention or an unintended Yes (1), No/Unclear (0)
exposure in both study groups?

Analytical and reporting bias

13 Were effect sizes based on the whole data or pre-defined subgroups rather than Yes (1), No/Unclear (0)
a post hoc portion of the data?
14 Were participants analyzed within the groups they were originally assigned to or, Yes (1), No/Unclear (0)
if not, was an intention-to-treat analysis performed?
15 Was a subset of the original recorded outcome variables or an incomplete set No (1), Yes/Unclear (0)
of key point estimates and variability measures reported?

An explanation and elaboration of these items can be found in Supplementary Appendix 2 (see online supplementary material at http://
www.liebertonline.com).
The total quality score ranges from 0 to 25 points.
SCI: spinal cord injury.
1784
quality of a study is closely intertwined with the quality of re-
porting;15,16 therefore, the devised quality scoring system required
each item to be both reported and indicative of a reduced suscep-
tibility to bias. The scoring system is applicable to both randomized
and nonrandomized controlled comparative surgical studies. Gui-
dance for scoring individual items is presented in an ‘‘Explanation
and Elaboration document,’’ which can be found in Supplementary
Appendix 2 (see online supplementary material at http://www
.liebertonline.com).
Outcome measures
Upon completion of the critical appraisal, outcome measures
suitable for meta-analysis were identified. The primary outcome
measure was neurological improvement, either measured via 1) the
mean difference in Total Motor Score as defined by the Interna-
tional Standards for Neurological Classification of Spinal Cord
Injury,17,18 or 2) the ‘‘neurological improvement rate’’ odds ratio
(OR); using different scales, including the American Spinal Injury
Association (ASIA) Impairment Scale17,18 and the Frankel scale.19
Secondary outcomes included the mean difference in length of
hospital stay, the mortality OR, and the adverse events OR.
Statistical analysis
All evaluated and reported effects of the timing of spinal surgery
after tSCI were abstracted and assessed for eligibility for meta-
analysis (Fig. 1). Outcomes’ point estimates and measures of vari-
ability (where applicable) had to be reported in at least three studies
in order to be considered for meta-analysis. For continuous out-
comes, for example, the Total Motor Score,17,18 we calculated the
weighted mean difference (WMD), that is, the weighted average of
the differences in the individual studies, the weight being the indi-
vidual inverse variances (i.e., precision) for each study adjusted for
heterogeneity (i.e., the variability in study effect sizes beyond that
caused by random error alone). This adjustment was based on as-
sessed study deficiencies with the QE model and simply on statistical
variability of effect sizes across studies with the random effects (RE)
model. Similarly, for binary outcomes, for example, occurrence of
nonspecific neurological improvements and adverse events, we
calculated the OR, again computing a weighted average across
studies as described. The advantage of the QE meta-analysis20,21 is
that it allows for redistribution of individual studies’ inverse variance
weights based on assessed study deficiencies rather than the usual
random redistribution of weights seen with the RE model.22 The
derived quality score of each individual study was normalized by
dividing by the maximum possible quality score and indicated as the
quality index (Qi, range: 0–1). Subsequently, Qi values were entered
in the quality effects model as described by Doi et al.20,21,23,24
Statistical heterogeneity was assessed using the I2 and s2 sta-
tistics. Heterogeneity was regarded as substantial when the I2 sta-
tistic exceeded 30% and the s2 was > 0. To assess the robustness of
the meta-analysis, we performed one-way and ‘‘criterion-based’’
sensitivity analyses. One-way sensitivity analysis was conducted
by excluding individual studies sequentially from the meta-analy-
ses. A criterion-based sensitivity analysis looks at the pooled results
after altering inclusion criteria for meta-analysis. Studies were
selected and grouped using the following predefined inclusion
criteria: 1) SCI severity, 2) level of SCI, and 3) maximum delay
‘‘early’’ treatment. The outcomes of a meta-analysis can be con-
sidered robust when sensitivity analyses in subsets of studies show
similar pooled effect sizes.
To test for small-study effects,25 or potential publication bias,
the symmetry of funnel plots was assessed visually and statistically
with the Egger’s linear regression test. Although appreciating other
possible causes of funnel plot asymmetry,26 publication bias was
suspected if the intercept of Egger’s regression line deviated from 0
with a p value of < 0$1.27 All statistical analyses were performed
VAN MIDDENDORP ET AL.
using MetaXL version 1.3 available from www.epigear.com.
During the reporting of current study’s findings we adhered to the
recommendations outlined in the Preferred Reporting Items for
Systematic Reviews and Meta-Analyses (PRISMA) statement.26,28
Results
Search and screening results
The computerized search strategy resulted in 2589 citations (Fig.
1). After screening of titles and abstracts, 42 remaining potential
eligible articles were obtained for full-text screening. Twenty-one
articles were excluded after a more detailed evaluation. The ma-
jority of the irrelevant articles did not evaluate the effect of the
timing of spinal surgery specifically, or compared operative and
nonoperative treatments. Cross-referencing resulted in one addi-
tional study. Twenty-two studies29–50 reported on the effects of
the timing spinal surgery after tSCI, of which 18 studies reported
usable information for meta-analysis.
Study characteristics and critical appraisal
Included studies consisted of 1 randomized controlled trial
(RCT) (5%), 1 quasi-RCT (5%), 4 prospective cohort studies
(18%), and 16 retrospective comparative cohort studies (73%)
(Table 2). Early treatment was defined as spinal surgery performed
within 24 and 72 h after injury in ten (46%) and six (27%) studies,
respectively. The maximum delay for ‘‘late’’ spinal surgery was 1
week in one study36 (5%), 2–8 weeks in four studies31,39,43,47
(18%), and 6–12 months in two studies33,40 (9%). A majority of 15
studies (68%) did not report the maximum delay for surgery.
Applied as a rough indicator for the level of SCI, half of the
studies (11) included only tetraplegics, whereas 5 studies (23%)
included only paraplegics. Thirteen studies (59%) included both
complete and incomplete SCI subjects, and seven studies (32%)
included only incomplete SCI subjects. Follow-up was notably
short for studies focusing on length of stay outcomes with a min-
imum reported mean duration of 10 days for early treated para-
plegics in one study.38 Studies that focused on neurological
outcomes reported longer follow-up periods ranging from 6 months
to 9 years. Only four studies (18%) reported sufficient details on
point estimates and measures of variability of the latest follow-up
visits.30,33,39,40
As indicated by the low Qi values presented in Table 2, included
studies demonstrated a high susceptibility to bias. A complete set of
assigned quality scores is presented in detail in Supplementary
Appendix 3 (see online supplementary material at http://www
.liebertonline.com).
Quantitative synthesis
Neurological outcomes were reported in 19 studies29,31–33,35–
37,39–50
(86%), length of stay related outcomes were reported in 12
studies30,31,34,37–39,41,42,45,47,48,50 (55%), financial outcomes were
reported in 2 studies34,41 (9%), bladder function outcomes were
reported in 2 studies33,37 (9%), and the Functional Independence
Measure51 and blood loss44 were each reported in 1 study (5%)
(Table 2).
Neurological outcomes
Nine studies29,32,33,37,41,42,45,48,50 (41%) reported the Total
Motor Score, six studies31,36,39,44,49,50 (27%) reported the neuro-
logical improvement rate, four studies31,36,41,50 (18%) reported the
 Table 2. Study Characteristics of Studies Meeting Inclusion Criteria

Included
Study Study Timing definitions subjects Mean age in SCI SCI Latest ‘‘protocol’’
Source design period (early/late)a Subgroups (early/late) yrs (early/ late)a Level Severity Outcomes follow-upa Qi
b
McAfee et al. 1985 RCS 1973 - 1981 NA - 48 30 (16–65) P I NEU 3.4 (2–8.6) yrs 0.08
Petitjean et al. 1995 RCS 1990–1993 < 24 h/ > 24 h - 10/19 37 – 17 P I,C NEU, LOS, NR 0.08
Other
Wiberg et al. 1988 RCS 1980–1985 < 24 h/ > 24 h, < 7 d/ > 7 d - 8/2/20 NR P I,C NEU (8–72) m 0.12
Anderson et al. 2012 RCS 2000–2009 < 24 h/ > 24 h, < 48 h/ > 48 h - 14/30/25 59 (23–89) T I NEU 11 (6–60) m 0.12
Clohisy et al. 1992 RCS 1981–1990 < 48 h/ > 48 h - 11/9 30 (15–63)/ 38 (17–66) P I NEU, Bladder 42 (12–62) m/42 0.16
(6–92) m
Campagnolo et al. RCS 1990–1996 < 24 h/ > 24 h - 37/27 32/ 42 T,P I,C LOS 38 – 34 d/55 – 40 d 0.16
1997
Pollard et al. 2003 RCS 1982–2000 < 24 h/ > 24 h -c 86/243 34 – 15/ 36 – 16 T C NEU, LOS NR 0.16
Sapkas et al. 2007 RCS 1987–2000 < 72 h/ > 72 h - 31/36 36 (16–72) T I,C NEU 4 (1–9) yrs 0.16
Chen et al.2009 RCS 1999–2004 £ 4 d/ > 4 d - 21/28 56 (22–76) T I NEU 56 (25-NR) m 0.16
Stevens et al. 2010 RCS 1985–2006 < 24 h/ > 24 h - 16/34 NR T I NEU, LOS NR 0.16
Duh et al. 1994 PCS 1985–1988 < 100 h/ > 100 h -c NR NR T,P I,C NEU 1 yr/1 yr 0.20
Newton et al. 2011 RCS 1988–2000 NA{ - 57 22 (14–47) T I,C NEU NR 0.20
Levi et al. 1991 PCS 1985–1990 < 24 h/ > 24 h I 10/40 30/33 T I,C NEU, LOS 9 (6–12) m/10 0.24

1785
(6–12) m
C 35/18 25/28 11 (9–12) m/12
(9–12) m
Mirza et al. 1999 RCS 1989–1991 < 72 h/ > 72 h - 15/15 32 (14–56)/33 (21–49) T I,C NEU, LOS NR 0.24
Croce et al. 2001 RCS 1996–1999 < 72 h/ > 72 h - 71/74 34/34 T,P NR LOS, Costs 29 d/36 d 0.24
Guest et al. 2002 RCS 1986–1996 < 24 h/ > 24 h - 16/34 41 – 20/48 – 20 T I NEU, LOS, 36 (13–48) m 0.24
Bladder
Kerwin et al. 2005 RCS 1988–2001 < 72 h/ > 72 h T 59/27 37/39 T,P NR LOS 21 d/24 d 0.24
P 31/15 38/38 10 d/31 d
McKinley et al. 2004 RCS 1995–2000 < 72 h/ > 4 d -c 307/296 37 – 15/36 – 16 T,P I,C NEU, LOS, 1 yr/1 yr 0.28
Costs, Other
Vaccaro et al. 1997 RCT 1992–1995 < 72 h/ > 5 d - 35/29 40 (15–75)/39 (16–75) T I,C NEU, LOS 10 m/ 3 m 0.32
Wilson et al. 2012 PCS 2007–2009 < 24 h/ > 24 h - 35/49 42/48 T,P I,C NEU, LOS NR 0.40
Fehlings et al. 2012 PCS 2002–2009 < 24 h/ > 24 h, £ 7 d - 182/131 45 – 17/51 – 16 T I,C NEU 6 m/6 m 0.52
Cengiz et al. 2008 Quasi- 2004–2006 < 8 h/ > 72 h, < 15d - 12/15 40 – 16/43 – 14 P I,C NEU, LOS 15 (12–20) m 0.56
RCT
a
A slash (‘/’) denotes details for both early and late groups were presented. Values in parentheses are ranges of mean values, when reported. bThe timing of surgical spinal decompression was processed as a
continuous measure in this study. cVarious subgroup analyses were presented in this article. However, as no details of the subgroups were presented these analyses were not included in the current review.
RCS, retrospective cohort study; PCS, prospective cohort study; RCT, randomized controlled trial; h: hour(s); T, tetraplegia; P, paraplegia; I, incomplete; C, complete; NEU, neurological; LOS, length of stay; Qi,
‘‘quality’’ index value (see Supplementary Appendix 3), NR, not reported; NA, not applicable.
1786 VAN MIDDENDORP ET AL.

FIG. 2. Forest plot of the Total Motor Score weighted mean differences (WMDs) in individual studies and pooled estimate using the
quality effects model.

ASIA Impairment Scale, and four studies42,46–48 (14%) reported the
Frankel scale. In addition, a variety of neurological aggregate
scores, sensory scores, and level of injury measures were each
reported in one or two studies.35,37,39,41,45 The Total Motor Score
and the neurological improvement rate outcomes were suitable
measures for meta-analysis.
The mean Total Motor Score improvement was specified for
both ‘‘early’’ and ‘‘late’’ groups in seven studies (32%), comprising
815 pooled subjects. Studies were considered heterogeneous
(s2 > 0; I2 = 52%; p = 0$05), and the QE model was used to pool the
data, with the RE model being used for standardized comparison.
The studies’ individual and pooled WMDs and 95% CIs obtained
with the QE model are shown in Figure 2. The pooled WMD of the
Total Motor Score between ‘‘early’’ and ‘‘late’’ surgical spinal
decompression was 5$94 (95% CI: 0.74, 11.15), indicating that, on
average, subjects who undergo early spinal surgery gain approxi-
mately six motor score points more than subjects who undergo late
treatment (Table 3). The RE model showed a slightly smaller
pooled WMD of 4$73 (95% CI: - 0$13, 9$59) points, with 0 now
being included in the CI. For details see Supplementary Appendix 4
(see online supplementary material at http://www.liebertonline
.com).
The neurological improvement rate was specified in six studies
(27%), comprising 495 pooled subjects. As a measure of

Table 3. Pooled Effects of Analyzed Outcomes Using Quality Effect, One-Way Sensitivity Analyses
and Random Effect Models

Studies/
Outcome and statistic comparisons Model Pooled effect Excluded study

Total Motor Score improvement 7/7 QE 5. 94 (0.74, 11.15) -

WMD QE-1WS: Early + 8.16 (1.61, 14.71) McKinley et al.2004
QE-1WS: Late + 3.62 ( - 1.15, 8.40) Mirza et al.1999
RE 4.73 ( - 0.13, 9.59) -
Neurological improvement rate 6/9 QE 2.23 (1.35, 3.67) -
OR QE-1WS: Early + 3.11 (1.50, 6.44) Fehlings et al. 2012
QE-1WS: Late + 1.60 (1.02, 2.50) Cengiz et al.2008
RE 1.74 (1.04, 2.91) -
Length of hospitalization (days) 6/6 QE - 9.98 ( - 13.10, - 6.85) -
WMD QE-1WS: Early + - 11.61 ( - 16.49, - 6.73) McKinley et al.2004
QE-1WS: Late + - 8.72 ( - 11.97, - 5.47) Cengiz et al.2008
RE - 8.51 ( - 12.78, - 4.25) -
Mortality 9/14 QE 0.97 (0.40, 2.31) -
OR QE-1WS: Early + 0.81 (0.31, 2.09) Fehlings et al. 2012 (FU)
QE-1WS: Late + 1.25 (0.48, 3.28) Kerwin et al. 2005
RE 0.70 (0.35, 1.41) -
Adverse events (all) 12/28 QE 0.71 (0.49, 1.04) -
OR RE 0.86 (0.69, 1.07) -

Values in parentheses are 95% confidence intervals.

WMD, weighted mean differences; OR, odds ratio; QE, quality effects; QE-1WS, quality effects one-way sensitivity analysis; RE, random effects;
Early + , exclusion of one study resulting in a pooled effect most favoring ‘‘early’’ spinal surgery, Late + , exclusion of one study resulting in a pooled
effect most favoring ‘‘late’’ spinal surgery.
TIMING OF SPINAL SURGERY AFTER tSCI
improvement, two studies applied one ASIA Impairment Scale
grade,36,50 one study applied one or more ASIA Impairment Scale
grades,31 one study applied one or more Frankel scale grades,49 and
two studies applied ‘‘any’’39 or ‘‘good’’44 improvement. Three
studies performed either two subgroup analyses39,44 or included
two ‘‘late’’ comparison groups,49 resulting in a total of nine com-
parisons for meta-analysis. The ORs were considered heteroge-
neous (s2 > 0; I2 = 15%; p = 0$31). The pooled OR for neurological
improvement was 2$23 (95% CI: 1$35, 3$76) indicating a twice as
high probability of improvement for subjects undergoing ‘‘early’’
intervention (Table 3). The RE model again showed a slightly
lower pooled OR of 1$74 (95% CI: 1$04, 2$91). Details of both
models are presented in Supplementary Appendix 5 (see online
supplementary material at http://www.liebertonline.com).
Length of stay outcomes
Twelve studies30,31,34,37–39,41,42,45,47,48,50 (55%) reported the
length of stay in hospital and/or rehabilitation center, six stud-
ies31,34,37,38,42,47 (27%) reported the length of stay in the intensive
care unit, and three studies34,38,42 (14%) reported the duration of
mechanical ventilation. The length of stay in hospital was the single
suitable measures for meta-analysis.
The mean length of hospital stay was specified in six studies
(27%), comprising 1103 pooled subjects. Studies were considered
heterogeneous (s2 > 0; I2 = 71%; p < 0$01). The pooled WMD of
the length of hospital stay was - 9$98 (95% CI: - 13$10, - 6$85),
indicating that on average, subjects who undergo early spinal sur-
gery spend 10 days less in hospital than do subjects who undergo
late treatment (Table 3 and Fig. 3). The RE model showed a smaller
pooled WMD of - 8$51 (95% CI: - 12$78, - 4$25). For details see
Supplementary Appendix 6 (see online supplementary material at
http://www.liebertonline.com).
Mortality
Out of the 10 studies29,31,34,36,38,39,46–49 (46%) that reported on
mortality outcomes, 9 were suitable for meta-analysis. Five studies
FIG. 3. Forest plot of the length of hospital stay weighted mean differences (WMDs, in days) in individual studies and pooled estimate
using the quality effects model.
1787
reported two comparisons: two studies evaluated two sub-
groups,38,39 two studies included two ‘‘late’’ comparison
groups,29,49 and one study evaluated short- and long-term follow-
up.36 This resulted in a total of 14 comparisons in the QE model,
comprising 1148 pooled subjects. The ORs were considered ho-
mogeneous (s2 = 0; I2 = 0%; p = 0$64). The pooled OR for mortality
was 0$97 (95% CI: 0$40, 2$31), indicating no differences in
mortality risks between subjects undergoing ‘‘early’’ and ‘‘late’’
spinal surgery (Table 3 and Fig. 4). The RE model showed a similar
pooled OR of 0$70 (95% CI: 0$35, 1$41). For details, see Sup-
plementary Appendix 7 (see online supplementary material at
http://www.liebertonline.com).
Other adverse events
Of the 19 studies (86%) reporting on adverse events, 5 studies
specified all reported adverse events in both groups (see Supple-
mentary Appendix 3). However, 12 studies (55%) reported at least
one adverse event in both groups and, therefore, were suitable for
meta-analysis. Twenty-eight comparisons were reported in the
following (post-hoc devised) subgroups: neurological deterioration
(five studies, 23%),36,37,42,45,49 thromboembolic events (two stud-
ies, 9%),36,41 (other) pulmonary events (four studies,
23%),31,34,38,41 sepsis (three studies, 14%),31,36,49 wound and skin
events (three studies, 14%),33,36,41 number of patients with adverse
events (two studies, 9%),39,47 and various other adverse events – six
in total – in four studies.33,36,37,41 Outcomes of analyses on all
adverse events combined are presented in Table 3.
No statistically significant differences were found for the sub-
groups using the QE model. However, the RE model demonstrated
a statistically significant pooled effect (OR: 0$67, 95% CI: 0$51,
0$89) for pulmonary events, indicating that subjects who under-
went ‘‘late’’ surgical intervention had a higher risk of developing
pulmonary complications, for example, pneumonia and atelectasis,
than did ‘‘early’’ treated subjects. However, caution is required
in interpretation of such random effect results, particularly
when considering that no significant effects were observed in the
quality effect model. Details of both models are presented in
1788
FIG. 4. Forest plot of mortality odds ratios (ORs) in individual studies and pooled estimate using the quality effects model.
Supplementary Appendix 8 (see online supplementary material at
http://www.liebertonline.com).
Sensitivity analysis and publication bias
One-way sensitivity analyses with each study individually re-
moved was done for all performed QE models, see Table 3. The
pooled estimate for Total Motor Score improvement varied con-
siderably after having excluded a single study resulting in a pooled
effect most (8$16, 95% CI: 1$61, 14$71)41 and least (3$62, 95%
CI: - 1$15, 8$40)42 favoring ‘‘early’’ spinal surgery. Similar effect
modifications were seen for the neurological improvement rate OR
analyses. Summary tables with data from the one way sensitivity
analyses are presented in Supplementary Appendix 9 (see online
supplementary material at http://www.liebertonline.com).
VAN MIDDENDORP ET AL.
Results of the ‘‘criterion-based’’ sensitivity analyses are pre-
sented in Table 4. Pooled estimates for the Total Motor Score
suggested smaller improvements for studies that included incom-
plete SCI subjects only, included both tetraplegics and paraplegics,
and – unexpectedly – considered ‘‘early’’ surgery as being within
24 h instead of later (Table 4). Summary estimates for the neuro-
logical improvement rate suggested smaller improvements for
studies that included incomplete SCI subjects or tetraplegics only.
Pooled estimates for length of hospital stay outcomes suggested
smaller differences for studies that included incomplete SCI sub-
jects only. Mortality sensitivity analyses demonstrated substan-
tially dispersed pooled risks for all criteria. The summary risk
estimate for adverse events was sensitive to change when tested for
severity of injury (incomplete SCI: higher risk) and timing of
‘‘early’’ surgery ( £ 24 h: lower risk, Table 4).
TIMING OF SPINAL SURGERY AFTER tSCI 1789

1.34 (0.38, 4.73)

0.65 (0.41, 1.02)

0.70 (0.42, 1.15)

0.68 (0.38, 1.20)

0.58 (0.34, 0.97)

0.90 (0.59, 1.38)

0.73 (0.18, 2.90)

0.71 (0.47, 1.07)
Presence of small-study effects was assessed by evaluating
funnel plots for symmetry both visually and statistically. A possi-
bility of publication bias was found for the following outcomes:

OR
events (all)
Adverse
Total Motor Score (intercept: 1$85, 95% CI: 0$66, 3$03; p = 0$01),
‘‘neurological improvement rate’’ (intercept: 0$9, 95% CI: - 0$05,
1$85; p = 0$06), and length of hospital stay (intercept: - 2$53, 95%
CI: - 4$76, - 0$29; p = 0$04) (Fig. 5a–c). The funnel plots clearly
illustrate that for all three outcomes it were the studies with smaller

5
21

13
16

16
13

6
23
n
samples that favored ‘‘early’’ spinal surgery.

1.07 (0.18, 6.49)

1.32 (0.37, 4.72)

0.90 (0.35, 2.33)

1.16 (0.19, 7.17)

1.32 (0.41, 4.29)

0.56 (0.16, 1.98)

0.94 (0.17, 5.08)

0.97 (0.36, 2.61)
Discussion
This meta-analysis demonstrated that patients who underwent

OR
Mortality
‘‘early’’ spinal surgery after tSCI had significantly greater neuro-

Table 4. Results of Criterion-Based Sensitivity Analyses Using a Quality Effects Model

logical recovery outcomes and significantly shorter hospital ad-
missions than did patients who underwent ‘‘late’’ spinal surgery.
However, this study also demonstrated that these findings are far
from robust, as estimates dispersed substantially in both one way

4b
7b

8
6

10
4

5
9
and criterion-based sensitivity analyses and funnel plots also

n
demonstrated a high likelihood of publication bias. Two major
sources of heterogeneity were identified: 1) patients with a variety

- 11.12 ( - 14.66, - 7.58)

- 8.70 ( - 13.97, - 3.43)

- 10.55 ( - 14.39, - 6.71)

- 10.96 ( - 16,36, - 5.56)

- 8.67 ( - 11.76, - 5.57)

- 8.80 ( - 15.75, - 1.86)

- 10.41 ( - 13.90, - 6.93)
of SCI severities and levels of injury were included in original

- 4.77 ( - 10.22, 0.68)

studies and 2) included studies demonstrated a high susceptibility

hospital stay (days)

to various sources of bias.
To our knowledge, this is the first literature review to present and

WMD
Length of
integrate both qualitative and quantitative data demonstrating a
profound heterogeneity among original studies reporting on the
effects of the timing of spinal surgery after tSCI. Although basic
methodological limitations have been addressed in previous re-
views and meta-analyses,5–10 no study has yet quantified the po-
tential impact of studies’ susceptibility to bias on pooled treatment

Median Quality index score = 0.20; btwo studies did not report the severity of the injury, see Table 2.
effect estimates. Previous reviews have addressed the methodo-

3
n

2
4

4
2

3
3
logical limitations of original studies, generally using an almost
instinctive, standard closing sentence referring to ‘‘the need for
3.86 (1.49, 10.03)

3.75 (0.57, 24.48)

randomized controlled trials.’’ However, clinical equipoise – 1.42 (0.88, 2.28)
1.35 (0.33, 5.44)
2.36 (1.35, 4.10)

2.13 (1.24, 3.66)

improvement rate

defined as a state of genuine uncertainty on the part of the clinical

Neurological

investigator regarding the comparative therapeutic merits of each

OR

-
-
arm in a trial52 – has lost ground, as a recent survey indicated an
outspoken preference for ‘‘early’’ spinal decompression among
spine surgeons.53 This essential aspect, combined with a number of

SCI, spinal cord injury; WMD, weight mean difference, OR, odds ratio.
other feasibility issues,54 makes conducting a properly powered,
multicenter, surgical RCT on tSCI patients a very challenging, if
3
n

2
7

9
0

4
5
not unfeasible, undertaking.54 The current study indicates that
much more can – and needs to be – done in addressing other sources
of heterogeneity than confounding by indication alone.
6.01 ( - 2.86, 14.88)

7.07 ( - 0.20, 14.34)

5.19 ( - 1.79, 12.17)

2.94 ( - 3.54, 9.43)

2.06 ( - 2.84, 6.95)

7.52 (0.44, 14.60)
5.54 (0.90, 10.18)

7.34 (0.92, 13.77)

The first identified source of heterogeneity relates to the variety

Values in parentheses are 95% confidence intervals.

of SCI severities and levels of injury included in original studies.

WMD
Total motor

SCI is a complex condition that presents with markedly heteroge-

score

neous manifestations related to the severity and anatomic level of

injury.55 Moreover, SCI patients may also present with concomi-
tant injuries and pre-existent comorbidities.56,57 Together with
age,58 these patient related factors have been demonstrated to be
strongly related to patient outcomes after tSCI.58–60 In the current
5
4
3

2
5

3
4

review, we found that 1) not a single study reported a complete

overview of patients’ age, SCI level, SCI severity, concomitant
Timing ‘‘early’’ group

injuries and co-morbidities; 2) only 2 of the 14 studies that included

Paraplegia w/wo
Tetraplegia only
Incomplete only
Complete w/wo

more than one type of severity and/or level of SCI, stratified sub-
Tetraplegia
Incomplete

Quality scorea

jects accordingly;38,39 and 3) only 4 of the 21 nonrandomized

£ 24 hours
> 24 hours

> median
SCI severity

£ median

studies (19%) applied statistical techniques to adjusted evaluated

SCI level
Criterion

outcome measures for confounding.35,36,40,50

Although beyond the scope of this review, it is worth mentioning
a

several countermeasures to address observed confounders in non-

1790 VAN MIDDENDORP ET AL.

FIG. 5. Funnel plots for Total Motor Score (a), neurological improvement rate (b), and length of hospital stay (c).
TIMING OF SPINAL SURGERY AFTER tSCI
randomized comparative studies, including matching, stratifica-
tion, and multivariate regression techniques.61 Statistical tech-
niques to adjust for unobserved, or unconsidered, confounders have
been documented, albeit such adjustments require a well-defined
high-volume cohort of patients.62 Finally, propensity scores can be
used to balance the covariates in the two study groups, and thus
reduce the magnitude of selection bias in nonrandomized studies.63
The second identified source of heterogeneity relates to the in-
dividual studies’ susceptibility to bias other than confounding by
indication. This was assessed using a tailored scoring system de-
vised based on items covered by three consensus-derived
‘‘checklists.’’12–14 Where these checklists mainly focus on the
detail of reporting, the devised scoring system was primarily in-
tended as a tool to assess individual, surgical studies’ susceptibility
to bias, or methodological quality. Some authors state that a study’s
true susceptibility to bias can only be assessed through its surrogate
measure: the quality of reporting.64 Although this argument makes
sense from a non-reporting perspective, it does not hold from a
‘‘do-reporting’’ perspective. To illustrate, author group ‘‘A’’ re-
ports that the ‘‘treatment allocation was random, and double-blind’’
and author group ‘‘B’’ reports that ‘‘treatment allocation was nei-
ther random, nor blind.’’ Both reports score a high ‘‘quality of
reporting’’ on this item, however, only group ‘‘A’’’s report scores a
high ‘‘methodological quality.’’ Nonetheless, most of the meth-
odological safeguards assessed in the current review were not re-
ported in original studies, and, therefore, may have resulted in an
overestimation of included studies’ true susceptibility to bias.
The specific sources of individual studies’ susceptibility to bias
are outlined in Supplementary Appendix 3. Although the six
prospective studies tended to present higher Qi values, the single
included RCT48 did not reach the highest quality score. Whereas
random treatment allocation is a key methodological safeguard,
other sources of heterogeneity than confounding by indication
alone may have an equally striking impact on studies’ validity. No
study performed and reported a sample size calculation, and only
two studies (9%) did report a primary outcome measure required
for determining a sample size. Whereas the manifestations of SCI
are highly variable, even more so are its recovery patterns. Sample
size calculations are of crucial importance in addressing the im-
pact of this disorder-specific variability on the outcomes of ther-
apeutic studies.59 In a recent protocol publication by the
Prospective, Observational European Multicenter study on the
efficacy of acute surgical decompression after traumatic Spinal
Cord Injury (SCI-POEM) investigators, a stratified sample size
calculation (by ASIA Impairment Scale grades) was reported and
based on reference data from a large European cohort of SCI
patients.54,58
The heterogeneity found in the current review may raise the
question of whether the eligibility criteria that were applied were
too loose. Tables 2 and 4 perfectly demonstrate, however, that if we
had applied more strict eligibility criteria for the sake of homoge-
neity, only a few studies would have been included, and the gen-
eralizability of our findings would have been limited. Discussed
sources of heterogeneity and susceptibility to bias had a profound
impact on the quantitative analyses presented in the current review.
However, the slight differences between the pooled effects and
variance reported in the RE and QE models may suggest otherwise.
The explanation for this is straightforward: as the Qi median value
of 0$20 indicates, a large majority of studies were susceptible to
multiple sources of bias. As such, quality discounting was not
limited to a few studies and resulted in an inevitable ‘‘quality ho-
mogeneity paradox.’’
1791
Although acknowledging the inherent limitations of applied
sensitivity and small-study effects analyses, several findings merit
further consideration. One-way sensitivity analyses indicated that
the pooled effects of both the Total Motor Score improvement and
the neurological improvement rate dispersed substantially after
excluding one single study from the analysis. In addition, criterion-
based sensitivity analyses demonstrated that none of the pooled
neurological outcome estimates were robust enough after stratify-
ing by each of the predefined criteria. These sensitivity analyses
clearly demonstrate the lack of robustness in these data and,
therefore, do not validate the statistically significant outcomes of
the current meta-analysis. No significant differences in mortality
were observed between the ‘‘early’’ and ‘‘late’’ treatment groups.
In the light of potential selection bias, one might expect to see a
higher mortality rate the ‘‘late’’ group. However, as only a few
studies reported the required information to explore this hypothesis,
no conclusive findings can be drawn from the current meta-
analysis. Finally, results from the funnel plots and related Egger
regression coefficients cannot be mistaken; there is a clear pre-
ponderance of smaller studies reporting effects favoring ‘‘early’’
spinal surgery. Without getting into a ‘‘chicken and egg’’ conun-
drum, the likelihood of observed publication bias coincides well
with recently reported spine surgeons’ outspoken preference for
‘‘early’’ spinal decompression after tSCI.53
No consensus exists as to what constitutes a clinically mean-
ingful neurological improvement. Where an improvement of five
Motor Score points may seem to be trivial for clinicians and re-
searchers, minimal neurological improvements could result in
profound functional benefits, particularly in tetraplegic patients.
However, only three studies related the clinical meaningfulness
of observed neurological improvements to functional outcomes,
that is, bladder function33,37 and the Functional Independence
Measure.51
Illustrated by the variety of applied definitions for ‘‘early’’ spinal
surgery in included studies, there is no consensus on what consti-
tutes the threshold value distinguishing ‘‘early’’ from ‘‘late’’ sur-
gery. More than 100 years after Burrell’s problem definition,1 it is
still unclear what the therapeutic window of opportunity after tSCI
is; and both clinical and pre-clinical scientists face a similar di-
lemma.65 The criterion-based sensitivity analyses demonstrated a
counterintuitively, greater neurological benefit in studies consid-
ering ‘‘early’’ surgery as within 72 h instead of within 24 h. This
raises another discussion about the limited reliability of very early
neurological examinations.66 Others have stated, however, that
when no concomitant injuries are present, and full cooperation can
be obtained, the timing of the examination itself is not affecting the
examination’s precision.67 Given that timing of surgery was con-
sidered as a ‘‘continuous’’ factor in the retrospective study reported
by Newton et al.43 (Qi: 0$20) it could not be included in the current
analysis. However, using receiver operating characteristics (ROC)
analysis, the latter authors determined that the optimal cutoff time
for surgical reduction based on the outcome ‘‘complete neurolog-
ical recovery after tSCI’’ was £ 4 h after injury for all patients as
well as for patients completely paralyzed on admission.43 Once
confirmed in future studies, this finding could have far-reaching
implications for pre-hospital management and logistics.
The variety of approaches for spinal surgery that has been ap-
plied within and between original studies has frequently been un-
derappreciated in previous literature reviews. The current review
demonstrated that the surgical approaches applied in both the
‘‘early’’ and ‘‘late’’ groups were reported and comparable in not
more than two studies (9%).31,39 In the remainder of the studies, it
1792
was not clear whether study groups had a comparable number of
patients subjected to closed, anterior, posterior, or circumferential
surgical approaches. In fact, 10 studies (46%) explicitly reported the
term ‘‘decompression,’’ or ‘‘reduction,’’ as part of the intervention
performed in all subjects. Fewer studies reported details on more
specific surgical techniques believed to play a role in the decom-
pression of spinal canal, including laminectomies and discectomies.
Only three studies (14%) indicated having performed postsurgical
imaging (magnetic resonance imaging in one study,36 computed
tomography in two studies31,33) for confirming a successful de-
compression of the spinal cord, although no criteria for successful
decompression were documented. Where pre-clinical scientists pay
utmost attention to the standardization of injury models,68 the current
review clearly uncovered clinical researchers’ limited attention to
technical, qualitative, and quantitative aspects of the actual surgical
decompression of the spinal cord in their final study reports.
Strengths of our study include independent screening and data
abstraction by two reviewers, the critical appraisal of studies’ sus-
ceptibility to bias, the inclusion of a variety of clinically relevant
outcomes, applied statistical techniques for assessing bias in this
review, and reporting according to the PRISMA statement’s rec-
ommendations.26 Nonetheless, several potential limitations of this
review exist. Although the tailored bias assessment scoring instru-
ment was based on items from established consensus-derived
‘‘checklists,’’12–14 no validation study has demonstrated its psy-
chometric properties. Pooling of outcome measures, including the
neurological improvement rate, may have resulted in additional
‘‘analytical heterogeneity.’’ Including two ‘‘late’’ surgery compari-
son groups in analyses led to doubling of ‘‘early’’ intervention group
data and may, therefore, have resulted in distorted outcomes. Sen-
sitivity analyses may have limited specificity, as the predefined cri-
teria were not adjusted. We did not perform a meta-regression
analysis, as the number of included studies was too small and the
overall quality of included studies was too low.26 Finally, studies that
did evaluate the effect of timing of surgical spinal decompression as
a continuous variable could not be included in meta-analysis.
Conclusion
In conclusion, despite the fact that ‘‘early’’ spinal surgery was
significantly associated with improved neurological and length of
stay outcomes, none of these positive associations were robust, as
demonstrated through sensitivity analyses. Moreover, funnel plots
showed significant proof of publication bias. Therefore, the au-
thors’ verdict is that because of different sources of heterogeneity
within and between original studies, there is no robust evidence
supporting ‘‘early’’ spinal surgery after tSCI. However, what evi-
dence there is clearly supports early intervention. Where properly
powered RCTs have proven to be an almost unfeasible study design
for acute, surgical tSCI management, we discussed several ap-
proaches that can be implemented in future nonrandomized studies,
and subsequently diminish their susceptibility to bias. By endorsing
such a high-quality standard, we may well find a definitive answer
to the hypotheses postulated by Burrell more than a century ago.
Acknowledgments
We thank Michael Schuetz for his contributions during the early
phases of this research project. Drs. van Middendorp and Doi
contributed to the design of this research project. Drs. van Mid-
dendorp and Hosman conducted the search and abstracted the data.
Dr. van Middendorp performed the statistical analysis. Dr. Doi
VAN MIDDENDORP ET AL.
assisted in data abstraction and statistical analysis. Dr. van Mid-
dendorp drafted the manuscript. All authors critically revised the
manuscript and approved its final version.
Author Disclosure Statement
No competing financial interests exist.
References
1. Burrell, H.L. (1905). Fracture of the spine. A summary of all the cases
(244) which were treated at the Boston City Hospital from 1864 to
1905. Ann. Surg. 42, 481–506.
2. Carlson, S.L., Parrish, M.E., Springer, J.E., Doty, K., and Dossett, L.
(1998). Acute inflammatory response in spinal cord following impact
injury. Exp. Neurol. 151, 77–88.
3. Dimar, J.R., 2nd, Glassman, S.D., Raque, G.H., Zhang, Y.P., and
Shields, C.B. (1999). The influence of spinal canal narrowing and
timing of decompression on neurologic recovery after spinal cord
contusion in a rat model. Spine 24, 1623–1633.
4. Rabinowitz, R.S., Eck, J.C., Harper, C.M., Jr., Larson, D.R., Jimenez,
M.A., Parisi, J.E., Friedman, J.A., Yaszemski, M.J., and Currier, B.L.
(2008). Urgent surgical decompression compared to methylpredniso-
lone for the treatment of acute spinal cord injury: a randomized pro-
spective study in beagle dogs. Spine (Phila Pa 1976) 33, 2260–2268.
5. Ball, J.R., and Sekhon, L.H. (2006). Timing of decompression and
fixation after spinal cord injury—when is surgery optimal? Crit. Care
Resusc. 8, 56–63.
6. Kishan, S., Vives, M.J., and Reiter, M.F. (2005). Timing of surgery
following spinal cord injury. J. Spinal Cord Med. 28, 11–19.
7. Schinkel, C., and Anastasiadis, A.P. (2008). The timing of spinal
stabilization in polytrauma and in patients with spinal cord injury.
Curr. Opin. Crit. Care 14, 685–689.
8. Fehlings, M.G., and Perrin, R.G. (2006). The timing of surgical in-
tervention in the treatment of spinal cord injury: a systematic review
of recent clinical evidence. Spine 31, S28–36.
9. Furlan, J.C., Noonan, V., Cadotte, D.W., and 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.
10. La Rosa, G., Conti, A., Cardali, S., Cacciola, F., and Tomasello, F.
(2004). Does early decompression improve neurological outcome of
spinal cord injured patients? Appraisal of the literature using a meta-
analytical approach. Spinal Cord 42, 503–512.
11. Lemeshow, A.R., Blum, R.E., Berlin, J.A., Stoto, M.A., and Colditz,
G.A. (2005). Searching one or two databases was insufficient for meta-
analysis of observational studies. J. Clin. Epidemiol. 58, 867–873.
12. Boutron, I., Moher, D., Altman, D.G., Schulz, K.F., and Ravaud, P.
(2008). Extending the CONSORT statement to randomized trials of
nonpharmacologic treatment: explanation and elaboration. Ann. In-
tern. Med. 148, 295–309.
13. van Tulder, M.W., Assendelft, W.J., Koes, B.W., and Bouter, L.M.
(1997). Method guidelines for systematic reviews in the Cochrane
Collaboration Back Review Group for Spinal Disorders. Spine (Phila
Pa 1976) 22, 2323–2330.
14. von Elm, E., Altman, D.G., Egger, M., Pocock, S.J., Gotzsche, P.C.,
and Vandenbroucke, J.P. (2007) Strengthening the Reporting of Ob-
servational Studies in Epidemiology (STROBE) statement: guidelines
for reporting observational studies. BMJ 335, 806–808.
15. Huwiler–Muntener, K., Juni, P., Junker, C., and Egger, M. (2002).
Quality of reporting of randomized trials as a measure of methodo-
logic quality. JAMA 287, 2801–2804.
16. Juni, P., Altman, D.G., and Egger, M. (2001). Systematic reviews in
health care: Assessing the quality of controlled clinical trials. BMJ
323, 42–46.
17. American Spinal Injury Association (Revised 2000). International
Standards for Neurological Classification 0f Spinal Cord Injury.
American Spinal Injury Association: Chicago.
18. Kirshblum, S.C., Burns, S.P., Biering–Sorensen, F., Donovan, W.,
Graves, D.E., Jha, A., Johansen, M., Jones, L., Krassioukov, A.,
Mulcahey, M.J., Schmidt–Read, M., and Waring, W. (2011). Inter-
national standards for neurological classification of spinal cord injury.
J. Spinal Cord Med. 34, 535–546.
19. Frankel, H.L., Hancock, D.O., Hyslop, G., Melzak, J., Michaelis, L.S.,
Ungar, G.H., Vernon, J.D., and Walsh, J.J. (1969). The value of
TIMING OF SPINAL SURGERY AFTER tSCI
postural reduction in the initial management of closed injuries of the
spine with paraplegia and tetraplegia. I. Paraplegia 7, 179–192.
20. Doi, S.A., and Thalib, L. (2008). A quality-effects model for meta-
analysis. Epidemiology 19, 94–100.
21. Doi, S.A., and Thalib, L. (2009). An alternative quality adjustor for the
quality effects model for meta-analysis. Epidemiology 20, 314.
22. Al Khalaf, M.M., Thalib, L., and Doi, S.A. (2011). Combining het-
erogenous studies using the random-effects model is a mistake and
leads to inconclusive meta-analyses. J. Clin. Epidemiol. 64, 119–123.
23. Doi, S.A., Barendregt, J.J., and Mozurkewich, E.L. (2012). Meta-
analysis of heterogeneous clinical trials: an empirical example. Con-
temp. Clin. Trials 32, 288–298.
24. Doi, S.A., Barendregt, J.J., and Onitilo, A.A. (2013). Methods for the
bias adjustment of meta-analyses of published observational studies. J.
Eval. Clin. Pract. 19, 653–657.
25. Sterne, J.A., Egger, M., and Smith, G.D. (2001). Systematic reviews in
health care: Investigating and dealing with publication and other
biases in meta-analysis. BMJ 323, 101–105.
26. Liberati, A., Altman, D.G., Tetzlaff, J., Mulrow, C., Gotzsche, P.C.,
Ioannidis, J.P., Clarke, M., Devereaux, P.J., Kleijnen, J., and Moher,
D. (2009). The PRISMA statement for reporting systematic reviews
and meta-analyses of studies that evaluate healthcare interventions:
explanation and elaboration. BMJ 339, b2700.
27. Egger, M., Davey Smith, G., Schneider, M., and Minder, C. (1997). Bias
in meta-analysis detected by a simple, graphical test. BMJ 315, 629–634.
28. Moher, D., Liberati, A., Tetzlaff, J., and Altman, D.G. (2009). Pre-
ferred reporting items for systematic reviews and meta-analyses: the
PRISMA statement. BMJ 339, b2535.
29. Anderson, D.G., Sayadipour, A., Limthongkul, W., Martin, N.D.,
Vaccaro, A., and Harrop, J.S. (2012). Traumatic central cord syn-
drome: neurologic recovery after surgical management. Am. J. Or-
thop. (Belle Mead NJ) 41, E104–108.
30. Campagnolo, D.I., Esquieres, R.E., and Kopacz, K.J. (1997) Effect of
timing of stabilization on length of stay and medical complications
following spinal cord injury. J. Spinal Cord. Med 20, 331–334.
31. Cengiz, S.L., Kalkan, E., Bayir, A., Ilik, K., and Basefer, A. (2008).
Timing of thoracolomber spine stabilization in trauma patients; impact
on neurological outcome and clinical course. A real prospective (rct)
randomized controlled study. Arch. Orthop. Trauma Surg. 128, 959–966.
32. Chen, L., Yang, H., Yang, T., Xu, Y., Bao, Z., and Tang T. (2009).
Effectiveness of surgical treatment for traumatic central cord syn-
drome. J. Neurosurg. Spine 10, 3–8.
33. Clohisy, J.C., Akbarnia, B.A., Bucholz, R.D., Burkus, J.K., and
Backer, R.J. (1992) Neurologic recovery associated with anterior de-
compression of spine fractures at the thoracolumbar junction (T12-
L1). Spine (Phila Pa 1976) 17, S325–330.
34. Croce, M.A., Bee, T.K., Pritchard, E., Miller, P.R., and Fabian, T.C.
(2001). Does optimal timing for spine fracture fixation exist? Ann.
Surg. 233, 851–858.
35. Duh, MS., Shepard, M.J., Wilberger, J.E., and Bracken, M.B. (1994) The
effectiveness of surgery on the treatment of acute spinal cord injury and its
relation to pharmacological treatment. Neurosurgery 35, 240–249.
36. Fehlings, M.G., Vaccaro, A., Wilson, J.R., Singh, A.D.W.C., Harrop,
J.S., Aarabi, B., Shaffrey, C., Dvorak, M., Fisher, C., Arnold, P.,
Massicotte, E.M., Lewis, S., and Rampersaud, R. (2012) Early versus
delayed decompression for traumatic cervical spinal cord injury: re-
sults of the Surgical Timing in Acute Spinal Cord Injury Study
(STASCIS). PLoS One 7, e32037.
37. Guest, J., Eleraky, M.A., Apostolides, P.J., Dickman, C.A., and
Sonntag, V.K. (2002). Traumatic central cord syndrome: results of
surgical management. J. Neurosurg. 97, 25–32.
38. Kerwin, A.J., Frykberg, E.R., Schinco, M.A., Griffen, M.M., Murphy,
T., and Tepas, J.J. (2005). The effect of early spine fixation on non-
neurologic outcome. J. Trauma 58, 15–21.
39. Levi, L., Wolf, A., Rigamonti, D., Ragheb, J., Mirvis, S., and Robinson,
W.L. (1991). Anterior decompression in cervical spine trauma: does the
timing of surgery affect the outcome? Neurosurgery 29, 216–222.
40. McAfee, P.C., Bohlman, H.H., and Yuan, H.A. (1985) Anterior de-
compression of traumatic thoracolumbar fractures with incomplete
neurological deficit using a retroperitoneal approach. J. Bone Joint
Surg. Am. 67, 89–104.
41. McKinley, W., Meade, M.A., Kirshblum, S., and Barnard, B. (2004).
Outcomes of early surgical management versus late or no surgical
intervention after acute spinal cord injury. Arch. Phys. Med. Rehabil.
85, 1818–1825.
1793
42. Mirza, S.K., Krengel, W.F., 3rd, Chapman, J.R., Anderson, P.A.,
Bailey, J.C., Grady, M.S., and Yuan, H.A. (1999). Early versus de-
layed surgery for acute cervical spinal cord injury. Clin. Orthop. Relat.
Res. 359, 104–114.
43. Newton, D., England, M., Doll, H., and Gardner, B.P. (2011). The case for
early treatment of dislocations of the cervical spine with cord involvement
sustained playing rugby. J. Bone Joint Surg. Br. 93, 1646–1652.
44. Petitjean, M.E., Mousselard, H., Pointillart, V., Lassie, P., Senegas, J.,
and Dabadie, P. (1995). Thoracic spinal trauma and associated in-
juries: should early spinal decompression be considered? J. Trauma
39, 368–372.
45. Pollard, M.E., and Apple, D.F. (2003). Factors associated with im-
proved neurologic outcomes in patients with incomplete tetraplegia.
Spine (Phila Pa 1976) 28, 33–39.
46. Sapkas, G.S., and Papadakis, S.A. (2007). Neurological outcome
following early versus delayed lower cervical spine surgery. J. Orthop.
Surg. (Hong Kong) 15, 183–186.
47. Stevens, E.A., Marsh, R., Wilson, J.A., Sweasey, T.A., Branch, C.L.,
Jr., and Powers, A.K. (2010). A review of surgical intervention in the
setting of traumatic central cord syndrome. Spine J 10, 874–880.
48. Vaccaro, A.R., Daugherty, R.J., Sheehan, T.P., Dante, S.J., Cotler,
J.M., Balderston, R.A., Herbison, G.J., and Northrup, B.E. (1997).
Neurologic outcome of early versus late surgery for cervical spinal
cord injury. Spine (Phila Pa 1976) 22, 2609–2613.
49. Wiberg, J., and Hauge, H.N. (1988). Neurological outcome after
surgery for thoracic and lumbar spine injuries. Acta Neurochir. (Wien)
91, 106–112.
50. Wilson, J.R., Singh, A., Craven, C., Verrier, M.C., Drew, B., Ahn, H.,
Ford, M., and Fehlings, M.G. (2012). Early versus late surgery for
traumatic spinal cord injury: the results of a prospective Canadian
cohort study. Spinal Cord 50, 840–843.
51. Keith, R.A., Granger, C.V., Hamilton, B.B., and Sherwin, F.S. (1987).
The functional independence measure: a new tool for rehabilitation.
Adv. Clin. Rehabil. 1, 6–18.
52. Freedman, B. (1987). Equipoise and the ethics of clinical research. N.
Engl. J. Med. 317, 141–145.
53. Fehlings, M.G., Rabin, D., Sears, W., Cadotte, D.W., and Aarabi, B.
(2010). Current practice in the timing of surgical intervention in spinal
cord injury. Spine (Phila Pa 1976) 35, S166–173.
54. van Middendorp, J.J., Barbagallo, G., Schuetz, M., and Hosman, A.J.
(2012). Design and rationale of a Prospective, Observational European
Multicenter study on the efficacy of acute surgical decompression
after traumatic Spinal Cord Injury: the SCI-POEM study. Spinal Cord
50, 686–694.
55. Tuszynski, M.H., Steeves, J.D., Fawcett, J.W., Lammertse, D., Ka-
lichman, M., Rask, C., Curt, A., Ditunno, J.F., Fehlings, M.G., Guest,
J.D., Ellaway, P.H., Kleitman, N., Bartlett, P.F., Blight, A.R., Dietz,
V., Dobkin, B.H., Grossman, R., and Privat, A. (2007). Guidelines for
the conduct of clinical trials for spinal cord injury as developed by the
ICCP Panel: clinical trial inclusion/exclusion criteria and ethics.
Spinal Cord 45, 222–231.
56. Furlan, J.C., Kattail, D., and Fehlings, M.G. (2009). The impact of co-
morbidities on age-related differences in mortality after acute trau-
matic spinal cord injury. J. Neurotrauma 26, 1361–1367.
57. Putz, C., Schuld, C., Akbar, M., Grieser, T., Wiedenhofer, B., Fur-
stenberg, C.H., Gerner, H.J., and Rupp, R. (2011) Neurological and
functional recovery in multiple injured patients with paraplegia: out-
come after 1 year. J. Trauma 70, 1078–1085.
58. van Middendorp, J.J., Hosman, A.J., Donders, A.R., Pouw, M.H.,
Ditunno, J.F., Jr., Curt, A., Geurts, A.C., and Van de Meent, H. (2011).
A clinical prediction rule for ambulation outcomes after traumatic
spinal cord injury: a longitudinal cohort study. Lancet 377, 1004–
1010.
59. Fawcett, J.W., Curt, A., Steeves, J.D., Coleman, W.P., Tuszynski,
M.H., Lammertse, D., Bartlett, P.F., Blight, A.R., Dietz, V., Ditunno,
J., Dobkin, B.H., Havton, L.A., Ellaway, P.H., Fehlings, M.G., Privat,
A., Grossman, R., Guest, J.D., Kleitman, N., Nakamura, M., Gaviria,
M., and Short, D. (2007). Guidelines for the conduct of clinical trials
for spinal cord injury as developed by the ICCP panel: spontaneous
recovery after spinal cord injury and statistical power needed for
therapeutic clinical trials. Spinal Cord 45, 190–205.
60. Kirshblum, S.C., Priebe, M.M., Ho, C.H., Scelza, W.M., Chiodo, A.E.,
and Wuermser, L.A. (2007) Spinal cord injury medicine. 3. Re-
habilitation phase after acute spinal cord injury. Arch. Phys. Med.
Rehabil. 88, S62–70.
1794 VAN MIDDENDORP ET AL.

61. Grimes, D.A., and Schulz, K.F. (2002) Bias and causal associations in 67. van Middendorp, J.J., Goss, B., Urquhart, S., Atresh, S., Williams,
observational research. Lancet 359, 248–252. R.P., and Schuetz, M. (2011). Diagnosis and prognosis of traumatic
62. Groenwold, R.H., Hak, E., and Hoes, A.W. (2009) Quantitative as- spinal cord injury. Global Spine J 1, 001–008.
sessment of unobserved confounding is mandatory in nonrandomized 68. Akhtar, A.Z., Pippin, J.J., and Sandusky, C.B. (2008). Animal models
intervention studies. J. Clin. Epidemiol. 62, 22–28. in spinal cord injury: a review. Rev. Neurosci. 19, 47–60.
63. D’Agostino, R.B., Jr. (1998). Propensity score methods for bias re-
duction in the comparison of a treatment to a non-randomized control
group. Stat. Med. 17, 2265–2281. Address correspondence to:
64. Karri, V. (2006). Randomised clinical trials in plastic surgery: survey Joost J. van Middendorp, MD, PhD, MClinEpid
of output and quality of reporting. J. Plast. Reconstr. Aesthet. Surg. 59, National Spinal Injuries Centre
787–796. Stoke Mandeville Hospital
65. Blesch, A., and Tuszynski, M.H. (2009) Spinal cord injury: plasticity,
regeneration and the challenge of translational drug development. Buckinghamshire Healthcare NHS Trust
Trends Neurosci. 32, 41–47. Aylesbury, Buckinghamshire, HP21 8AL
66. Kirshblum, S.C., and O’Connor, K.C. (1998). Predicting neurologic United Kingdom
recovery in traumatic cervical spinal cord injury. Arch. Phys. Med.
Rehabil. 79, 1456–1466. E-mail: jvanmiddendorp@gmail.com