Archives of Physical Medicine and Rehabilitation

journal homepage: www.archives-pmr.org

Archives of Physical Medicine and Rehabilitation 2015;96:133-40

REVIEW ARTICLE (META-ANALYSES)

Prevalence of Depression After Spinal Cord Injury:

A Meta-Analysis
Ryan Williams, PhD,a Adrian Murray, MAb
From the aAmerican Institutes for Research, Chicago, IL; and bDepartment of Counseling, Educational Psychology, and Research, College of
Education, Health and Human Sciences, University of Memphis, Memphis, TN.

Abstract
Objectives: To use meta-analysis to synthesize point prevalence estimates of depressive disorder diagnoses for persons who have sustained a
spinal cord injury (SCI).
Data Sources: We searched PsycINFO, PubMed, the Cumulative Index to Nursing and Allied Health Literature (CINAHL), and Dissertation
Abstracts International (DAI) for studies examining depression after SCI through 2013. We also conducted a manual search of the reference
sections of included studies.
Study Selection: Included studies contained persons with SCI; used a diagnostic measure of depression (ie, an unstructured, semi-structured, or
structured clinical interview, and/or a clinician diagnosis); and provided a diagnosis of major or minor depressive episodes for the subjects in the
study. Diagnostic criteria were based on the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, or the Diagnostic and
Statistical Manual of Mental Disorders-Third Edition (including Research Diagnostic Criteria) criteria.
Data Extraction: The 2 authors of this study screened the titles and abstracts of 1053 unique studies for inclusion in this meta-analysis. Nineteen
studies, containing 35,676 subjects and 21 effect size estimates, were included.
Data Synthesis: The mean prevalence estimate of depression diagnosis after SCI was 22.2%, with a lower-bound estimate of 18.7% and an upper
bound estimate of 26.3%. Random effects and mixed effects models were used in this work. A small number of study moderators were explored,
including sample sex composition, Diagnostic and Statistical Manual of Mental Disorders version used, data collection method (primary vs
secondary), sample traumatic etiology composition, sample injury level and completeness composition, and sample diagnostic composition. Data
collection method, Diagnostic and Statistical Manual of Mental Disorders version, and diagnostic composition significantly predicted variation in
observed effect size estimates, with primary data collection studies having lower estimates compared with secondary data analysis studies, studies
using Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, diagnostic criteria having higher estimates compared with studies
using Diagnostic and Statistical Manual of Mental Disorders, Third Edition, criteria, and samples comprising individuals diagnosed only with
major depression having lower prevalence estimates.
Conclusions: The existing data on depression after SCI indicate that the prevalence of depression after SCI is substantially greater than that in the
general medical population. These results underscore the importance of continued research on measuring depression in persons with SCI and on
treatments for depression after SCI.
Archives of Physical Medicine and Rehabilitation 2015;96:133-40
ª 2015 by the American Congress of Rehabilitation Medicine

Depression inhibits the physical rehabilitation process1 and exacer-
bates physical health problems associated with spinal cord injury
(SCI).2 Typical physical complications expected after SCI include
impaired respiratory function and an elevated risk for pneumonia and
atelectasis, neurogenic bowel dysfunction, gallstones, neurogenic
bladder, urinary tract infections, sexual dysfunction, pressure ulcers,
Disclosures: none.
thrombophlebitis, spasticity, heterotopic ossification, osteoporosis,
and chronic pain.3 Depressive disorders compound the debilitating
effects of these physical complications.
SCI may occur acutely (eg, from a traumatic event such as a
fall or car accident) or chronically (eg, from illness such as
spinal tumors, transverse myelitis, or Guillan-Barré syndrome).
The distinction between the 2 etiologies is important. Traumatic
SCI is an unanticipated negative life-course event, and it is ex-
pected to have a more profound psychological effect, at least

0003-9993/14/$36 - see front matter ª 2015 by the American Congress of Rehabilitation Medicine
http://dx.doi.org/10.1016/j.apmr.2014.08.016
134
immediately, for the person. Chronic syndromes are important to
consider for life-course development and psychopathology;
however, physiological transformation for many chronic condi-
tions tends to occur more slowly. Perhaps for this reason, most
research on psychological aspects of SCI focuses on trau-
matic SCI.
Estimates of depression prevalence rates vary dramatically
from study to study. The general understanding is that rates of
depressive disorders after SCI are higher than the prevalence es-
timates of the general medical population, but most studies about
depression after SCI inevitably state a very wide range of possible
population values. For example, in their review of the literature,
Bombardier et al4 found that major depression prevalence esti-
mates appeared to range from 9.8% to 38%.
Most of what we know about the prevalence of depressive
disorders after SCI is also based on self-report measures. These
self-report measures have infrequently been validated against
diagnostic criteria. In one such example, Bombardier et al5
compared a range of cutoff scores on the Patient Health
Questionnaire-9 (PHQ-9) to a diagnosis of major depression using
the Structured Clinical Interview for Diagnostic and Statistical
Manual of Mental Disorders, Fourth Edition (DSM-IV).6 The
PHQ-9 is one of the most popular depression measures used, and
the authors found that a total cutoff score of
11 provided
optimal sensitivity and specificity, both >.80. In that sample,
about 24.6% of the individuals screened positive for “probable
major depression.”
With such wide variance in depression prevalence estimates
among persons with SCI, researchers and practitioners should
be concerned about potential sources of variation, which may
not be solely due to sampling variability. One such source of
variation could be the sample’s demographic characteristics
such as sex, age, and race. Another source could be the diag-
nostic standard that the prevalence estimate was based on, as
well as the measure used to produce the depression scores
originally. Variability in the medical characteristics of these
samples, such as time since injury and level and completeness of
injury, may further explain the wide variation in depression
prevalence estimates.
To date, there have been no systematic efforts to quantitatively
synthesize the existing research on depression prevalence in per-
sons with SCI that also explores potential sources of variability
in those prevalence estimates. Kalpakjian et al7 mention that
combining various classes of depression measures, such as severity,
screening, and diagnostic, may provide an inaccurate and poten-
tially more confusing result. Systematically searching for and
quantitatively synthesizing depression prevalence estimates from
studies that formally diagnosed depression after SCI may provide
List of abbreviations:
CI confidence interval
DSM Diagnostic and Statistical Manual of Mental Disorders
DSM-III Diagnostic and Statistical Manual of Mental Disorders,
Third Edition
DSM-III-R Diagnostic and Statistical Manual of Mental Disorders,
Revised Third Edition
DSM-IV Diagnostic and Statistical Manual of Mental Disorders,
Fourth Edition
MDD major depressive disorder
PHQ-9 Patient Health Questionnaire-9
RDC Research Diagnostic Criteria
SCI spinal cord injury
R. Williams, A. Murray
an optimally useful and efficient estimate. Such an estimate could
clarify the specific risk of depression among individuals with
SCI and help practitioners allocate their time and screening efforts
effectively.
The purpose of this study was to estimate the prevalence of
depression diagnoses after SCI. The focus was on studies that used
diagnostic measures to identify depressive disorders. A systematic
literature review was conducted and random effects and mixed
effects models were fit to the data. Exploratory moderator ana-
lyses were used to evaluate potential sources of effect size het-
erogeneity, including sample’s sex composition, Diagnostic and
Statistical Manual of Mental Disorders (DSM ) version used for
diagnosis, data collection method (primary data collection vs
secondary data collection), sample composition of traumatic in-
juries, and sample composition of injury level and completeness.
The following section describes the details of the search, coding,
and analysis in accordance with the Preferred Reporting Items for
Systematic Review and Meta-Analyses.
Methods
Literature search and information retrieval
A systematic search of the literature was undertaken. PsycINFO,
PubMed, the Cumulative Index to Nursing and Allied Health
Literature (CINAHL), and Dissertation Abstracts International
(DAI) were searched. Using Boolean operators, we used the
wildcard term depress* in combination with each of the following
to isolate the population of studies investigating depression after
SCI: spinal cord injury, spinal cord injuries, spinal injury, spinal
injuries, spine injury, and spine injuries. The search included all
relevant records through 2013. We also manually searched the
reference section of each article meeting or preliminarily meeting
our inclusion criteria after the screening process.
Titles and abstracts were screened. Studies were retained if
their titles or abstracts indicated that depression was measured. If
abstracts indicated depression severity rather than diagnosis, the
study was excluded. If it was unclear whether a measure was used
for diagnostic purposes, it was retained for full-text screening.
Studies were included in the analysis if they included a diag-
nostic measure of depression (ie, an unstructured, semi-
structured, or structured clinical interview and/or a clinician
diagnosis); if they provided a diagnosis of major or minor
depressive episodes for the subjects in the study; if the sample
consisted of persons with SCI (traumatic or a mix of traumatic
and nontraumatic); and if the diagnostic criteria used in the study
were based on, or were exactly, the DSM-IV or the Diagnostic
and Statistical Manual of Mental Disorders, Third Edition (DSM-
III ) (including Research Diagnostic Criteria [RDC]) diagnostic
criteria. The 2 authors of this article searched for and screened
the studies in this review.
Study coding
Characteristics of the studies and their samples were coded for this
analysis. Study-level characteristics used in this analysis included
year of publication, study location, study setting, data collection
method, and publication source. Sample characteristics included
sex composition, racial composition, sample age, injury level
composition, injury completeness composition, traumatic injury
composition, and time since injury. We also coded sample size,
diagnostic measure used, and diagnostic criteria used.
www.archives-pmr.org
Prevalence of depression after SCI
Statistical procedures
Inverse-variance weighted effect size methods8 were used in this
meta-analysis. The proportion of individuals diagnosed with a
depressive disorder after SCI was the effect size estimate for this
study. Under some circumstances, the raw study proportion can
artificially inflate the estimated between-study heterogeneity and
underestimate variability in the mean proportion, namely, when
values are very small (<.20) or very large (>.80).9 Because many
of the estimates in this study were close to or below .20, the raw
proportions were transformed into logits, or log odds. The method
for this transformation is presented along with the SE estimator
and inverse variance weight estimator in Equations 1 to 3:


pk
ESk Zlog ; ð1Þ
1  pbk
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 1
sk Z þ ; ð2Þ
nk pk nk ð1  pk Þ
1 sex-based features, as well as the bereavement exclusion, which
wk Z 2 ; ð3Þ
sk
where p is the kth raw study proportion of individuals with a
depression diagnosis, n is the kth study sample size, ES is the log
odds of depression diagnosis for study k, and s is the kth study SE
estimate. All analyses were conducted on the logits, but the mean
effect size estimate was transformed back into the original pro-
portion metric using Equation 4:
eESk
pk Z : ð4Þ
1 þ ESk
After estimating a mean effect (ie, mean prevalence estimate),
moderator effects were explored using meta-regression. We
examined the following moderators: data collection method (pri-
mary data collection vs medical records review), DSM version
used for diagnosis (DSM-III and RDC vs DSM-IV), diagnosis
(major depressive disorder [MDD]-only samples vs mixed MDD
and non-MDD depression diagnosis samples), sample proportion
of individuals with a traumatic injury, sample proportion of in-
dividuals with tetraplegia, and sample proportion of men.
Diagnostic criteria have evolved over the past for decades and
that is why we included DSM version as a moderator in our an-
alyses, which include studies using the following criteria: RDC,
DSM-III, Diagnostic and Statistical Manual of Mental Disorders,
Revised Third Edition (DSM-III-R), and DSM-IV criteria.
RDC defines a depressive disorder as sustained and pervasive
dysphoric mood accompanied by biological, behavioral, and
cognitive symptoms.9 As clinical knowledge of depression
increased, more specific criteria were available for use in both
service- and research-oriented contexts. DSM-III and DSM-III-R
criteria for MDD are very similar. Both specify that dysphoric
mood or lack of interest/pleasure in all or almost all usual activ-
ities must be present and relatively persistent for a period of at
least 2 weeks, accompanied by at least 4 of the following symp-
toms: significant changes in appetite or weight; insomnia or
hypersomnia; psychomotor agitation or retardation; loss of energy
or fatigue; feelings of worthlessness or excessive guilt; complaints
or evidence of diminished ability to think or concentrate; and
recurrent thoughts of death, suicidal ideation, or suicide attempt.11
DSM-III-R improves on these criteria by providing an operational
definition of relatively persistent stating that the primary
www.archives-pmr.org
135
symptoms “occur most of the day, nearly every day” for the
2-week period. In addition, DSM-III-R provides more text on the
growing clinical knowledge surrounding depression (average age
at onset, course, impairment, complications, and predisposing
factors) and the use of clinical specifiers such as mild, moderate,
severe, in full or partial remission, unspecified, melancholic type,
and seasonal pattern.12
A notable change to these criteria occurred with the intro-
duction of DSM-IV. Specifically, a diagnosis of MDD must be
accompanied by clinically significant distress or impairment in
social, occupational, or other important areas of functioning. In
addition, the text provides information on variability in presenta-
tion, for instance, allowing the clinician to infer depressed mood
from facial expression and demeanor when the individual com-
plains of feeling anxious or having no feelings. Similarly, it is
noted that an emphasis on somatic complaints may be identified as
a substitute for reporting feelings of sadness. Irritability, previ-
ously discussed only as a facet of depression in children, was also
identified as a characteristic of depression in adults in this version
of the DSM. Finally, more text was added regarding cultural and
states that a diagnosis of depression is not made when the
depressive symptoms are better accounted for by bereavement.13
As with DSM-III-R, Diagnostic and Statistical Manual of
Mental Disorders, Fourth Edition, Text Revision, improves on its
preceding version by adding more information on the prevalence,
physical findings, and familial patterns associated with depression,
as well as the following specifiers: catatonic features, atypical
features, and postpartum onset.14
This information is important to the present study because the
rate of depression in a sample at any given time is not only a
function of the true variation in diagnoses but also dependent on
the understanding of depression at the time of diagnosis, the
criteria used, and the quality of information on other disorders,
allowing for a more thorough differential diagnosis.
Random effects models were used in this study. The random
effects variance component was computed using the method of
moments estimator.15 Heterogeneity was assessed using t2, the
between-studies variance component, and the I2 statistic, which is
the percentage of the total variability among studies. All statistical
analyses were conducted using R,16 including the R pack-
age metafor.17
We conducted a graphical analysis to investigate publication
bias using a funnel plot, which plots the observed effect size es-
timates against the inverse of their SEs. The effects used in this
study were unrelated to primary research questions in the primary
study, and they were provided outside a null hypothesis signifi-
cance testing framework. That is, the primary studies included in
this meta-analysis were not concerned with identifying the sta-
tistical significance of their sample’s depression prevalence esti-
mates. Because of this, we did not expect publication or outcome
reporting bias to operate in the usual way (ie, omission related to
statistical significance). It is possible, however, that effects rele-
vant to this study were censored because they were part of studies
with primary outcomes failing to meet statistical significance.
Unfortunately, there are no known meta-analytic tools to handle
such a situation.
Three of the included studies used self-report measures with
diagnostic criteria cutoff scores to diagnose depressive disor-
ders.18-20 These studies were included because their depression
measures contained a discrete cutoff score in which a DSM
depression diagnosis was made, rather than “probable depression.”
136 R. Williams, A. Murray

However, because these measures may function differently than
the clinical diagnostic measures used in the other included studies,
we conducted a sensitivity analysis by excluding these 3 studies
and reestimating the mean prevalence estimate.
One study21 was nearly half the size of the total meta-analytic
sample, with 17,565 individuals. Because of its size and potential
influence in this analysis, we conducted a sensitivity analysis of
the mean prevalence estimate by excluding it and comparing the
results to the overall analysis.
Another study22 estimated depression prevalence at multiple
time points: initial hospitalization, 3 months postinjury, and 6
months postinjury. Two other studies10,23 appear to use part of a
common sample, building on an earlier study. Because we used
each of these estimates in our analyses, we used robust variance
estimation24-26 to adjust our SEs for the correlation among effect
(ie, dependencies). For these analyses, we specifically used the
robust variance estimator with a small-sample correction,27 which
has proven to be an effective method for adjusting for effect size
dependencies while retaining appropriate type I error rates. Robust
variance estimation facilitates estimation when effect size esti-
mates are correlated. Specifically, this approach is useful because
it does not require any additional information about the effect
size estimates or their variances. Without accounting for such
dependencies, the SEs around the mean estimator or any moder-
ator coefficients may be inappropriately small, leading to inac-
curate inference.
Results
Information retrieval
We identified 1976 citations from our systematic literature search.
After identifying duplicates among the returned citations, 1053
unique articles were screened, by title and abstract, for inclusion.
Ultimately 22 studies met our inclusion criteria, although 3 of
these studies did not report the necessary information to compute
the relevant effect size estimates. Nineteen studies, with a total of
21 effect size estimates, were used in this meta-analysis. Two of
these studies were doctoral dissertations. Details about these
studies are provided below.
Sample characteristics
Table 1 reports basic descriptive information on the 19 studies
included in this analysis, including study authors, publication year,
sample size, percentage of men, percentage of individuals with

Table 1 Sample descriptive statistics

Diagnostic Method Mean Depression
Study Sample Sample Size Men (%) Tetraplegia (%) (Diagnosis) Age (y) SCI Type Diagnosis (%)
Fullerton et al (1981)23 30 90.00 50.00 SADS (MDD)* 27.5 Traumatic SCI 30.00
Howell et al (1981)10 22 86.40 100.00 SADS (minor 22.7 Traumatic SCI 22.70
depression)*
Frank et al (1985)28 32 ND 71.90 SSI (MDD)y 27 Traumatic SCI 37.50
Judd et al (1986)29 84 84.50 38.10 SSI (MDD)y 35.5 Traumatic SCI 7.00
MacDonald et al (1987)20 53 77.40 45.50 CDM (MDD)y 34.8 Traumatic SCI 13.00
Federoff et al (1991)30 55 ND ND PSE (mixed)y 33 Traumatic SCI 22.00
Frank et al (1992)19 134 97.80 ND IDD (MDD)y 39.8 Traumatic SCI 13.00
Judd and Brown (1992)31 227 81.90 40.00 SSI (MDD)y 34.2 Traumatic SCI 13.70
Tate et al (1993)32 30 ND 70.00 SSI (MDD)y 31.3 Traumatic SCI 23.00
Kishi et al (1994)22 60 86.70 27.00 Modified PSE (MDD)y ND Traumatic SCI 21.70
Kishi et al (1994)22 44 81.82 30.00 Modified PSE (MDD)y ND Traumatic SCI 22.73
Kishi et al (1994)22 31 77.42 23.00 Modified PSE (MDD)y ND Traumatic SCI 12.90
Clay et al (1995)18 133 97.70 40.00 IDD (MDD)y 39.9 Traumatic SCI 13.50
Smith (2004)21,z 17,656 98.40 43.00 MR/ICD-9 (mixed)x 54.1 Traumatic SCI 20.60
and disease
Dryden et al (2005)33 201 77.10 28.00 MR/ICD-9 (mixed)x 31 Traumatic SCI 28.90
Reed (2008)34,z 133 95.50 42.00 MR/ICD-9 (mixed)x 43.82 Traumatic SCI 27.30
and disease
Banerjea et al (2009)35 8338 97.90 44.00 MR/ICD-9 (mixed)x ND Traumatic SCI 26.70
Findley et al (2011)36 8334 98.10 43.00 MR/ICD-9 (mixed)x ND Traumatic SCI 26.20
Rabadi and Vincent (2011)37 87 97.70 51.00 MR/ICD-9 (mixed)x 60.0k Traumatic SCI 47.70
Weeks et al (2011)38 67 61.20 ND MR/ICD-9 (mixed)x 65.5 Traumatic SCI 15.00
Bombardier et al (2012)5 142 78.20 67.00 SCID (MDD)x 42.2 Traumatic SCI 10.00
Abbreviations: CDM, Clinical Depression Measure; IDD, Inventory to Diagnose Depression; MR/ICD-9, Medical Records/International Classification of
Diseases, Ninth Revision; ND, no data; PSE, Present State Exam; SADS, Schedule of Affective Disorders; SCID, Structured Clinical Interview for DSM
Disorders; SSI, semi-structured clinical interview.
* DSM-IV criteria.
y
Research diagnostic criteria.
z
Doctoral dissertation.
x
DSM-III criteria.
k
Median age.

www.archives-pmr.org
Prevalence of depression after SCI
tetraplegia, diagnostic method used, diagnostic criteria used, mean
age, SCI type (ie, traumatic SCI sample or traumatic SCI and
spinal cord disease sample), and percentage diagnosed with
depression. Several studies did not provide information on all the
characteristics listed in this table. The included studies spanned 31
years and represented a total of 35,818 persons with SCI. The
largest sample in this pool was 17,656,21 which was pulled from a
large military veterans database, and the smallest sample was 22
individuals.10 Each of the samples predominantly consisted of
men, and the mean age ranged from about 23 to about 66 years.
The percentage of individuals with tetraplegia varied quite a bit,
with a range of about 23% to 100%. Most of the studies focused
solely on MDD diagnoses. Several studies, especially those that
used medical records, had samples that contained multiple
depressive disorder diagnoses such as minor depression, dysthy-
mia, and MDD. One study had a small sample of individuals who
met criteria only for minor depression.10 In studies in which a
single depressive disorder was not specifically mentioned, we
assumed that the diagnoses in the sample were mixed.
Nine studies in this analysis used DSM-III or DSM-III-R
criteria or measures based on these criteria, 8 used International
Classification of Diseases, Ninth Revision, codes corresponding
to DSM-IV criteria, and 2 used RDC. In early 1980, RDC was
commonly used before the availability of DSM-III and it was used
in the 2 earliest studies included in this meta-analysis (both
published in 1981).
Fig 1 Forest plot with random effects mean depression prevalence estimate.
www.archives-pmr.org
137
Mean depression prevalence estimate
Overall, the mean prevalence estimate of depression after SCI was
22.2% (95% confidence interval [CI], 18.7%e26.3%). Figure 1
presents a plot of the sample effect size estimates, their fixed ef-
fect 95% CIs, and the random effects mean with a 95% CI. These
effects were substantially heterogeneous (t2 Z .04, I2 Z 90.41).
The mean effect was largely unchanged when the 3 studies
using self-report diagnostic methods were removed; the mean
prevalence estimate was 23.1% (95% CI, 19.5%e27.2%). The
mean depression prevalence estimate after excluding Smith,21
which had an exceptionally large sample, was 21% (95% CI,
18.1%e24.8%).
Moderator analyses
The results of the exploratory moderator analyses are presented
in table 2. Three of the examined moderator variables were sta-
tistically significant. The odds of a depression diagnosis in studies
using primary data collection were less than the odds of a
depression diagnosis in studies that extracted secondary data from
medical chart reviews (bZ1.03; P<.05). The rate of depression
diagnosis was significantly less among studies using DSM-IV
diagnostic criteria than among those that used DSM-III or RDC
criteria (bZ.47; P<.05). And, the odds of diagnosis were signif-
icantly less among studies that included only MDD diagnoses
than among those that included MDD and other depressive
138 R. Williams, A. Murray

Table 2 Moderator results, log odds coefficients

Moderator Variable Log-Odds Estimate (Odds) SE t Value df I2 t2
Data collection method (primary data collection vs 0.66* (0.52) 0.19 3.44 13.6 88.79 0.03
medical chart review)
DSM version (DSM-IV vs DSM-III/RDC) 0.52* (1.68) 0.20 2.59 13.69 89.49 0.03
Diagnostic composition (MDD only vs minor depression/mixed) 0.70* (0.49) 0.19 3.70 12.70 88.54 0.03
Proportion men 2.06 (7.84) 1.27 1.62 5.48 92.09 0.04
SCI type (traumatic SCI vs traumatic SCI and spinal disease) 0.17 (1.18) 0.24 0.69 1.49 84.39 0.05
Proportion tetraplegia 0.11 (1.11) 0.96 0.11 4.18 91.88 0.04
* Statistically significant at the .05 level.

diagnoses (bZ.66; P<.05). The moderator analyses for sex Discussion

composition and injury composition were heavily underpowered
because of missing data on these 2 variables. Importantly, sub-
stantial between-study residual heterogeneity was observed for
each of the moderator analyses.
Publication bias
Publication bias was examined using a graphical analysis of a
funnel plot, the results of which are displayed in figure 2. The
effects are plotted against the inverse of the sample SEs. Ideally,
the display should show a symmetrical funnel shape with fewer
effects associated with larger inverse SEs (and larger sample sizes).
Although the funnel plot is a limited method for investigating the
effects of publication or outcome-reporting bias, the effects in this
study are largely symmetrically distributed about the mean effect.
The mechanisms through which outcome-reporting bias or publi-
cation bias generally operate (ie, statistical significance) were not
focal to the effects used in the samples of this meta-analysis,
consistent with the findings of the funnel plot display. It is
possible, however, that some studies were never published because
of other reasons such as small sample size or null findings in other
aspects of the study (eg, primary outcome results).
Historically, estimates of the prevalence of depression after SCI
have varied substantially, leading to an unclear understanding of
how common this issue is and to what degree resources are
necessary to both identify and treat depression in this population.
This study conducted a systematic review and synthesized the
extant published and unpublished research on the prevalence of
depressive disorder diagnoses after SCI. Across 19 studies, the
mean depression prevalence estimate was about 22% (95% CI,
19%e26%), a finding not readily apparent from a cursory review
of existing research, which has a range of about 7% to about 47%.
Depression is a commonly occurring disorder, with about 16%
of the general medical population in the United States meeting
criteria for MDD at least once in their lifetime.39 Twelve-month
prevalence estimates are about 6.7%.40 The results of this study
further validate the notion that individuals with SCI are at a
greater risk of depression.
In addition, the results of this meta-analysis indicate that
depression diagnoses varied with several study characteristics.
Prevalence was lower in primary data collection studies than in
secondary data collection studies. Primary data collection studies
largely consist of participants actively seeking treatment for

Fig 2 Funnel plot of effect size estimates around mean effect.

www.archives-pmr.org
Prevalence of depression after SCI
various complications associated with their injury. Accordingly,
they may have lower rates of depression than do those individuals
who were part of secondary data collections studies. Prevalence
estimates were also smaller in samples in which MDD was
diagnosed than in samples that diagnosed a mix of depressive
disorders such as MDD and minor depression, which is likely the
result of diagnosing multiple conditions and inflating the numer-
ator in the prevalence estimate in those mixed samples. And in
studies that used DSM-IV criteria for depression diagnosis,
depression prevalence estimates were higher than in those that
used criteria from DSM-III or RDC (the DSM-III predecessor). As
the DSM is revised, one intention is that its diagnostic sensitivity
and specificity are also improved, which may partially explain this
finding. This relation should be further studied as studies of
depression after SCI using Diagnostic and Statistical Manual of
Mental Disorders, Fifth Edition, criteria emerge.
Clinically, improving the accuracy of depression prevalence
estimates after SCI can improve clinicians’ understanding of the
nature of depression after SCI, which can then be used to improve
screening and treatment efforts. Knowing the degree to which
depression is experienced by individuals in this population may also
assist in normalizing the diagnosis for individuals who experience
traumatic SCI. These results further underscore the importance of
research in measuring depression in those with SCI, with diagnostic
sensitivity, as well as identifying important correlates of and treat-
ments for depression after SCI. As investigations in this area increase
in scope, future research could examine protective factors that
decrease the likelihood of developing depression post-SCI such as
the tendency to seek social support and engage with difficult
emotional experiences as opposed to isolation and avoidance. Un-
fortunately, reporting of sample characteristics limited our ability to
thoroughly investigate other sources of effect size heterogeneity.
Study limitations
This meta-analysis was limited to studies using formal diagnostic
methods and diagnostic criteria. Other studies on depression after
SCI have focused attention on “probable depression” in the absence
of formal diagnostic measures. Only recently have researchers
begun examining the sensitivity of more commonly used self-report
measures of depression severity such as the PHQ-9 against diag-
nostic measures such as the Structured Clinical Interview for DSM
Disorders4 in persons with SCI. It is likely now that future updates of
this meta-analysis will be able to effectively use information from
studies using these emerging severity scores or cutoff scores for
“probable depression” on measures such as the PHQ-9.
Another important limitation was the limited availability of
important study-level moderators such as time since injury. Un-
fortunately, and unexpectedly, studies that otherwise met our in-
clusion criteria rarely reported time since injury in their sample
characteristics. This is an important moderator that has implica-
tions for treatment course. Fortunately, most contemporary studies
record and report this information and as this literature matures,
we will have future opportunities to examine the relation between
time since injury and prevalence.
Conclusions
Depression after SCI can compound physiological complications
related to the injury. As such, having a stable and accurate un-
derstanding of the prevalence of this phenomenon is critical. This
meta-analysis provides a comprehensive search and robust anal-
ysis of the existing literature on the prevalence of depressive
www.archives-pmr.org
139
disorders after SCI, providing a more specific understanding of the
nature of depression after SCI. It is clear that these estimates vary
substantially from study to study for a number of reasons. That
heterogeneity is reflected in the mean prevalence estimate syn-
thesized in this study, which also substantially improves the
generalizability of the findings. Researchers and practitioners may
use this information to improve screening.
Keywords
Depression; Meta-analysis; Prevalence; Rehabilitation; Review,
systematic; Spinal cord injuries
Corresponding author
Ryan Williams, PhD, American Institutes for Research, 20 North
Wacker Dr, Ste 1231, Chicago, IL 60606. E-mail address:
rwilliams@air.org.
References
1. Aoki T, Hosaka T, Ishida A. Psychiatric evaluation of physical reha-
bilitation patients. Gen Hosp Psychiatry 1995;17:440-3.
2. Elliott TR, Frank RG. Depression following spinal cord injury. Arch
Phys Med Rehabil 1996;77:816-23.
3. Sipski ML, Richards JS. Spinal cord injury rehabilitation: state of the
science. Am J Phys Med Rehabil 2006;85:310-42.
4. Bombardier CH, Richards JS, Krause JS, Tulsky D, Tate DG. Symp-
toms of major depression in people with spinal cord injury: implica-
tions for screening. Arch Phys Med Rehabil 2004;85:1749-56.
5. Bombardier CH, Fann JR, Tate DG, et al. An exploration of modifiable
risk factors for depression after spinal cord injury: which factors
should we target? Arch Phys Med Rehabil 2012;93:775-81.
6. First MB, Spitzer RL, Gibbon M, Williams BW. Structured Clinical
Interview for DSM-IV-TR Axis I Disorders, Research Version, Non-
Patient Edition (SCID-I/NP). New York: New York State Psychiatric
Institute, Biometrics Research; 2002.
7. Kalpakjian CZ, Bombardier CH, Schomer K, Brown PA, Johnson KL.
Measuring depression in persons with spinal cord injury: a systematic
review. J Spinal Cord Med 2009;32:6-24.
8. Hedges LV, Olkin I, Statistiker M, Olkin I, Olkin I. Statistical
methods for meta-analysis. New York: Academic Pr; 1985. Available
at: http://oldjll.sustainabilityforhealth.org/trial_records/20th_Century/
1980s/hedges/hedges-kp.pdf. Accessed February 7, 2014.
9. Lipsey MW, Wilson D. Practical meta-analysis. 2000. Available at:
http://www.citeulike.org/group/302/article/7081810. Accessed February
7, 2014.
10. Howell T, Fullerton DT, Harvey RF, Klein M. Depression in spinal
cord injured patients. Spinal Cord 1981;19:284-8.
11. American Psychiatric Association. Diagnostic and statistical manual
of mental disorders. 3rd ed. Washington (DC): American Psychiatric
Association; 1980.
12. American Psychiatric Association. Diagnostic and statistical manual
of mental disorders. 3rd ed., text rev. Washington (DC): American
Psychiatric Association; 1987.
13. American Psychiatric Association. Diagnostic and statistical manual
of mental disorders. 4th ed. Washington (DC): American Psychiatric
Association; 1994.
14. American Psychiatric Association. Diagnostic and statistical manual
of mental disorders. 4th ed, text rev. Washington (DC): American
Psychiatric Association; 2000.
15. DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin
Trials 1986;7:177-88.
140
16. R Core Development Team. R: a language and environment for statistical
computing. Vienna, Austria: R Foundation for Statistical Computing; 2013.
17. Viechtbauer W. Conducting meta-analyses in R with the metafor
package. J Stat Softw 2010;36:1-48.
18. Clay DL, Hagglund KJ, Frank RG, Elliott TR, Chaney JM. Enhancing
the accuracy of depression diagnosis in patients with spinal cord injury
using Bayesian analysis. Rehabil Psychol 1995;40:171.
19. Frank RG, Chaney JM, Clay DL, et al. Dysphoria: a major symptom
factor in persons with disability or chronic illness. Psychiatry Res
1992;43:231-41.
20. MacDonald MR, Nielson WR, Cameron MG. Depression and activity
patterns of spinal cord injured persons living in the community. Arch
Phys Med Rehabil 1987;68:339.
21. Smith B. The management of depression in persons with SCI&D:
identification of depression, use of antidepressants, and impact on out-
comes in SCI&D. 2004. Available at: http://search.proquest.com.ezproxy.
memphis.edu/pqdt/docview/305205247/abstract/45244707047B493APQ/
1?accountidZ14582. Accessed February 7, 2014.
22. Kishi Y, Robinson RG, Forrester AW. Prospective longitudinal study
of depression following spinal cord injury. J Neuropsychiatry Clin
Neurosci 1994;6:237-44.
23. Fullerton DT, Harvey RF, Klein MH, Howell T. Psychiatric disorders in
patients with spinal cord injuries. Arch Gen Psychiatry 1981;38:1369.
24. Williams RT. Using robust standard errors to combine multiple regres-
sion estimates with meta-analysis. 2012. Available at: http://search.
proquest.com.ezproxy.memphis.edu/pqdt/docview/1081489864/abstract/
D1B2BFB288284CEFPQ/1?accountidZ14582. Accessed February
7, 2014.
25. Hedges LV, Tipton E, Johnson MC. Robust variance estimation in
meta-regression with dependent effect size estimates. Res Synth
Methods 2010;1:39-65.
26. Tipton E. Robust variance estimation in meta-regression with binary
dependent effects. Res Synth Methods. Available at: http://onlinelibrary.
wiley.com/doi/10.1002/jrsm.1070/full. Accessed February 8, 2014.
27. Tipton E. Small sample adjustments for robust variance estimation with
meta-regression. Psychol Methods 2014 Apr 28 [Epub ahead of print].
28. Frank RG, Kashani JH, Wonderlich SA, Lising A, Visot LR.
Depression and adrenal function in spinal cord injury. Am J Psychiatry
1985;142:252-3.
R. Williams, A. Murray
29. Judd FK, Burrows GD, Brown DJ. Depression following acute spinal
cord injury. Spinal Cord 1986;24:358-63.
30. Fedoroff JP, Lipsey JR, Starkstein SE, Forrester A, Price TR,
Robinson RG. Phenomenological comparisons of major depression
following stroke, myocardial infarction or spinal cord lesions. J Affect
Disord 1991;22:83-9.
31. Judd FK, Brown DJ. Psychiatric consultation in a spinal injuries unit.
Aust N Z J Psychiatry 1992;26:218-22.
32. Tate DG, Forchheimer M, Maynard F, Davidoff G, Dijkers M.
Comparing two measures of depression in spinal cord injury. Rehabil
Psychol 1993;38:53.
33. Dryden DM, Saunders LD, Rowe BH, et al. Depression following
traumatic spinal cord injury. Neuroepidemiology 2005;25:55-61.
34. Reed K. Evaluation of the Quick Inventory of Depressive
Symptomatology-Self Report (QIDS-SR) in a spinal cord injury popu-
lation. 2008. Available at: http://search.proquest.com.ezproxy.memphis.
edu/pqdt/docview/304556626/abstract/96A88B89E8BF4E9BPQ/1?
accountidZ14582. Accessed February 7, 2014.
35. Banerjea R, Findley PA, Smith B, Findley T, Sambamoorthi U. Co-
occurring medical and mental illness and substance use disorders
among veteran clinic users with spinal cord injury patients with
complexities. Spinal Cord 2009;47:789-95.
36. Findley PA, Banerjea R, Sambamoorthi U. Excess mortality asso-
ciated with mental illness and substance use disorders among vet-
eran clinic users with spinal cord injury. Disabil Rehabil 2011;33:
1608-15.
37. Rabadi MH, Vincent AS. Do vascular risk factors contribute to the
prevalence of pressure ulcer in veterans with spinal cord injury? J
Spinal Cord Med 2011;34:46.
38. Weeks DL, Greer CL, Bray BS, Schwartz CR, White JR Jr. Associ-
ation of antidepressant medication therapy with inpatient rehabilita-
tion outcomes for stroke, traumatic brain injury, or traumatic spinal
cord injury. Arch Phys Med Rehabil 2011;92:683-95.
39. Kesler RC, Berglund P, Demler O, et al. The epidemiology of major
depressive disorder: results from the National Comorbidity Survey
Replication (NCS-R). JAMA 2003;289:3095-105.
40. Kessler RC, Chiu W, Demler O, Walters E. Prevalence, severity,
and comorbidity of 12-month DSM-IV disorders in the National Co-
morbidity Survey Replication. Arch Gen Psychiatry 2005;62:617-27.
www.archives-pmr.org