The British Journal of Psychiatry (2018)

Page 1 of 8. doi: 10.1192/bjp.2018.27

Review article

Depression and hopelessness as risk factors

for suicide ideation, attempts and death:
meta-analysis of longitudinal studies
Jessica D. Ribeiro, Xieyining Huang, Kathryn R. Fox and Joseph C. Franklin

Background Results
Many studies have documented robust relationships between Overall prediction was weaker than anticipated, with weighted
depression and hopelessness and subsequent suicidal thoughts mean odds ratios of 1.96 (1.81–2.13) for ideation, 1.63 (1.55–1.72)
and behaviours; however, much weaker and non-significant for attempt and 1.33 (1.18–1.49) for death. Adjusting for publi-
effects have also been reported. These inconsistencies raise cation bias further reduced estimates. Effects generally per-
questions about whether and to what degree these factors sisted regardless of sample severity, sample age or follow-up
confer risk for suicidal thoughts and behaviours. length.

Conclusions
Aims Several methodological constraints were prominent across
This study aimed to evaluate the magnitude and clinical utility of studies; addressing these issues would likely be fruitful moving
depression and hopelessness as risk factors for suicide ideation, forward.
attempts and death.
Declaration of interest
None.
Method
We conducted a meta-analysis of published studies from 1971 to Keywords
31 December 2014 that included at least one longitudinal ana- Suicide; risk factors; depression; hopelessness; meta-analysis
lysis predicting suicide ideation, attempt or death using any
depression or hopelessness variable. Copyright and usage
© The Royal College of Psychiatrists 2018.

Background repeatedly reported depression as the most common mental

illness among suicide decedents.12,13
As one of the leading causes of death worldwide, suicide accounts
Consistent with theories and practice recommendations, many
for nearly one million deaths each year. Suicide attempts are more
prospective studies have found that depression and hopelessness do
frequent still.1,2 The scope and gravity of this public health
confer substantial risk for later suicide ideation,14,15 attempts16,17
problem has prompted substantial increases in research. Despite
and death,17,18 with some studies even reporting 20- to 30-fold
this, suicide rates have not abated.3
increases in risk (for example Beck et al19 Juon et al20). Yet, many
Effective prevention rests on accurate prediction. Accordingly,
other studies report substantially weaker21–23 or even non-signifi-
identifying the risk factors for suicidal thoughts and behaviours
cant effects.24–27 These inconsistencies raise questions about
has been a major focus of suicide research. Risk factors are longitu-
whether depression and hopelessness indeed confer risk for future
dinal predictors of an outcome that divide a population into low-
suicidal thoughts and behaviours, and, if so, to what degree.
and high-risk groups. These are distinct from correlates, which
are associated with an outcome but lack evidence of temporal pre-
cedence; as such, correlates are not very useful in prediction Objectives
efforts.4 Although cross-sectional studies can identify correlates, The objective of this study is to address these questions. To this end,
establishing risk factors requires longitudinal designs. we conducted a quantitative meta-analysis of all longitudinal studies
Unipolar depression and hopelessness are among the most examining the prospective effects of depression and hopelessness on
commonly cited risk factors for suicidal thoughts and behaviours. suicide ideation, attempts and death. We address four aims. First, we
As evidence of this, depression and hopelessness are consistently provide a descriptive summary of the existing longitudinal litera-
featured within risk-factor guidelines developed by major national ture. Second, we examine the predictive power of depression and
and international organisations and commonly integrated into hopelessness as risk factors for future suicide ideation, attempts
structured suicide-risk assessments. In fact, the US Preventive and death. Third, we evaluate the potential moderating effects of
Task Force5 recommendations suggest that suicide-risk assessments sample severity, sample age and study follow-up length, as variabil-
only be conducted in the context of depression screenings. ity in these methodological factors may influence the effects of
Supporting these practices, a number of prominent suicide theories depression and hopelessness. Fourth, we consider the clinical
incorporate hopelessness and depression as either necessary (for utility of these results by examining effects in terms of the absolute
example Interpersonal–Psychological Theory of Suicide6,7) or suffi- risk of our outcomes (i.e. suicide ideation, attempt and death).
cient (for example Hopelessness Theory of Suicide8,9) causes of sui- Beyond depression per se, a large body of literature also exists
cidal thoughts and behaviours. Moreover, suicidal thoughts and exploring the effects of factors commonly studied in the context
behaviours are symptoms of a depressive episode,10 large epidemio- of depression (such as social isolation, rumination) on suicidal
logical studies have linked non-fatal suicide attempts with the pres- thoughts and behaviours; although slightly outside the scope of
ence of a mood disorder11 and psychological autopsy studies have the present investigation, we include analyses examining the

1
Ribeiro et al

effects of these factors within the supplementary material (See example suicide, suicidality, suicidal behavior, suicidal, suicide
Supplementary Table 1, available at https://doi.org/10.1192/bjp. death, suicide plan, suicide thoughts, suicide ideation, suicide
2018.27). gesture, suicide threat, non-suicidal self-injury, self-injury, self-
Of note, the present study represents an extension and specifi- harm, self-mutilation, deliberate self-harm, self-cutting, cutting,
cation of a larger meta-analysis conducted by Franklin et al,28 self-burning, self-poisoning). We also searched the reference sec-
which examined broad trends in suicide prediction since the incep- tions of all papers identified through database searches.
tion of suicide research. The approach applied by Franklin and
colleagues28 was too coarse to speak to the effects of specific con-
Study selection, inclusion criteria
structs or potentially important moderation effects for specific con-
structs. As depression and hopelessness have played pivotal roles in Inclusion criteria required that papers include at least one longitu-
suicide research as well as clinical policy and practice, it is prudent dinal analysis predicting suicide ideation, attempts or death using
to further investigate the effects of these constructs and the potential any variable related to depression or hopelessness. Papers were
moderators of their effects. One possibility is that, consistent with required to be peer-reviewed publications published in English.
Franklin et al,28 analyses reveal weak effects of depression and hope- We elected to use only published studies as we wanted to summarise
lessness. An alternate possibility, however, is that results of finer- the literature used to inform research, policy and practice and
grained analyses indicate stronger effects of these constructs, relying only on published studies provides some safeguards for
which would be more consistent with existing theories, policies study quality. Also, unpublished studies located could fail to be a
and practice. This may be possible particularly in the context of complete or representative sample of unpublished data, which can
moderation analyses. Results would stand to have important impli- in turn bias results. Papers were excluded if:
cations moving forward, and these implications may diverge from (a) no longitudinal analyses were reported
those delineated by Franklin et al.28 As such, the present study (b) no analyses examined the discrete effects on suicide ideation,
was designed to address this existing gap in knowledge. attempts or death (as we were interested in understanding the
specificity of effects on discrete suicide-relevant outcomes, out-
comes that combined suicidal thoughts and behaviours and/or
Method were not specific to suicidal thoughts or behaviours were not
considered as outcomes);
Sources and searches (c) analyses were conducted within a primary treatment study.
We conducted literature searches using PubMed, PsycINFO and No other criteria were used to exclude studies. Our initial
Google Scholar up to 31 December 2014. As search terms, we searches yielded 2541 peer-reviewed publications. Based on
used variants of the terms longitudinal (for example longitudinal, abstracts, 2372 did not meet inclusion criteria. The remaining arti-
predicts, prediction, prospective, future, later) and suicide (for cles were read in full; 166 studies were retained (see Fig. 1 for

2541
papers identified

Each
Excluded 1,822 papers that clearly did not
abstract
meet inclusion criteria
reviewed

719
papers screened-in

Each
553 papers excluded because did not meet
full-text article
inclusion criteria
reviewed

166
papers included
Included all unique cases where a variable
related to depression or hopelessness was
Each
used to longitudinally predict suicide
statistical test
ideation, attempts, or death
reviewed
(i.e. ‘prediction case’)

357
prediction cases
included

Fig. 1 PRISMA diagram.

2
 Depression and hopelessness as risk factors for suicide

PRISMA flow chart and Supplementary Data 1 for the references for tables, correlations, independent group means and group sample
the studies included in the meta-analysis.). size and t-statistics or P-value. As hazard ratios cannot be converted
into ORs, they were analysed separately (see Supplementary Data 2).
Extraction and coding Outliers were defined as values greater than 5 s.d. from the mean (n =
3, 0.01%). Results were nearly identical when outliers were retained.
Any statistical test that used a variable related to depression or hope- To ensure reliability of effects, we required a minimum of four predic-
lessness to predict suicide ideation, attempt or death was retained tion cases to report an estimate.
for analysis and termed a ‘prediction case’. To ensure case inde- Between-study heterogeneity was quantified using I2-tests.
pendence, we removed duplicate cases (n = 7), which occurred Between-study variance is common across studies because of differ-
when the same data were re-analysed across multiple publications ences in methodology; as such, a random-effects model was used for
and/or multiple follow-up assessments using the same predictors all meta-analyses. Random-effects models assume a distribution of
were included in a single study. In the first case, we retained the effects across studies. Thus, the combined effect estimated using a
most inclusive study; in the latter, we retained estimates at the random-effects model represents the mean of the distribution of
final assessment. Our rationale for doing so was that the furthest true effects rather than a single true effect (as in fixed-effects
time point was often most inclusive; however, results remained vir- models). Given that between-study variance is common, fixed-
tually unchanged when we retained the shortest time point instead. effects models are rarely indicated. Using random-effects models,
We identified a total of 357 unique prediction cases. heterogeneity across studies is accounted for in the weighting and
Data extracted from each study included: authors, publication calculation of each prediction case.
year, follow-up length, number of participants with histories of To quantify publication bias, we considered multiple indices.
self-injurious thoughts or behaviours, sample type (general popula- Specifically, we evaluated Duval and Tweedie’s trim and fill test,
tion (i.e. samples in which no participant was selected on the basis of Egger’s regression test, Begg and Mazumdar rank correlation test,
psychopathology or self-injurious thoughts or behaviour history), Classic fail-safe N, Orwin’s fail-safe N and funnel plot symmetry.
clinical (i.e. samples in which participants were selected on the In this meta-analysis, we assumed case independence. However,
basis of psychopathology history), self-injurious (i.e. samples in some studies contributed multiple prediction cases for which some
which participants were selected on the basis of self-injurious of the predictors may have been correlated (for example number of
thoughts or behaviour history), sample age group (i.e. adult, adoles- episodes; symptom severity). Simulation studies suggest that not
cent, mixed)), predictor variable, outcome variable and any relevant accounting for existing dependence has very little effect on point esti-
statistics from each prediction case. mates, but can result in slight underestimates of variance and confi-
The process for selecting and coding prediction cases involved dence intervals, which increases type 1 error.29 Fully accounting for
several iterative steps to ensure the validity and completeness of dependence requires modelling the covariance structure of each
our meta-analysis. Each prediction case was assigned risk-factor cat- case; however, published studies rarely reported this information.
egories to ensure meaningful summary of effects. Initial codes were Given this, we examined the potential effects of dependence by
independently determined by one of the authors with an advanced running analyses that assumed complete dependence of cases within
degree in psychology (i.e. masters or doctoral level). Each code was each study (compared with Scammacca et al30). Although these ana-
subsequently examined by two additional authors with advanced lyses represent a level of dependence that far exceeds the actual levels,
degrees in psychology. Discrepancies were resolved through discus- they provide insight into the upper limit of the effects of dependence
sion until consensus. on our results. Due to space limitations, we provide only results from
these models for overall prediction. Results from these analyses were
Study quality highly consistent with those assuming complete independence.
Methodological variability across studies may influence the accuracy For moderator analyses, we employed meta-regression using a
of results in meta-analyses. The inclusion criteria in the present meta- random-effects model and unrestricted maximum likelihood. In
analysis, however, required that studies share a common core design meta-analyses, moderator analyses test whether the variation in
(i.e. longitudinal) and outcome (i.e. suicide ideation, attempt or effect size across studies is associated with differences in modera-
death), thereby substantially constraining the study pool. As a tors. In this meta-analysis, we were interested in the potential mod-
result, relative to typical treatment meta-analyses, this meta-analysis erating effects of sample age group, sample severity and study
includes studies that are highly methodologically uniform. follow-up length. We examined these potential moderation effects
Nonetheless, finer-grained methodological differences between on overall prediction as well as on the effects of depression and
studies exist; however, no criteria have been established to definitely hopelessness.
inform a priori hypotheses about how these differences may have an
impact on study quality. For example, no objective guidelines exist Results
about how length of follow-up (for example 1 month v. 1 decade) or
sample severity (for example general population v. self-injuring) Descriptive summary of existing literature
influence prediction. As such, to empirically evaluate the effects of
these methodological differences, we conducted moderator ana- Publication dates ranged from 1971 to 2014. The number of pre-
lyses. We also accounted for study heterogeneity by using diction cases has sharply increased over time, with 2.0% of cases
random-effects models. published before 1985, 14.1% published in 1984–1994, 32.5% in
1995–2004 and 51.4% in 2005–2014. Suicide attempt was the
most common outcome (47.7%) followed by death (33.6%) and
Statistical analyses ideation (18.7%). The majority of cases used adult samples
Meta-analyses were performed using Comprehensive Meta-Analysis, (69.8%); 23.4% used adolescent samples and 6.8% used mixed
Version 3.29 When provided, zero-order (i.e. unadjusted) effects were adult–adolescent samples. Among adolescent samples, the most
used as this provided the purest effect estimate (results were highly common outcome was attempt (63.9%) followed by ideation
consistent across analyses using only unadjusted estimates v. only (33.7%); death cases were rare (2.4%).
adjusted v. both unadjusted and adjusted). When odds ratios (ORs) Only 16.4% of cases used self-injurious samples; clinical
were not provided, we calculated them based on 2 × 2 contingency (52.5%) and general (31.1%) samples were much more common.

3
Ribeiro et al

Clinical samples were the most common sample type within attempt (wOR = 2.38, 95% CI 1.84–3.07) and death (wOR = 1.50,
suicide death (60.5%) and attempt (56.8%) cases, followed by 95% CI 1.04– 2.17).
general (death: 17.6%; attempt: 27.2%) and then self-injurious Depressive symptoms were significant predictors of ideation
samples (death: 21.8%; attempt: 16.0%). Among ideation prediction (wOR = 1.57, 95% CI 1.45–1.70) and attempt (wOR = 1.30, 95%
cases, general samples were predominant (65.2%), and clinical CI 1.23–1.37) but not death. Several self-report inventories were
(27.3%) and self-injurious (7.6%) were less common. particularly common, allowing for additional analyses examining
The average follow-up length was about 9 years (median 60 the effects of particular self-reports. The Beck Depression
months; s.d. = 122.20 months; range: 1–708 months). Death by Inventory (BDI) and Center for Epidemiologic Studies Depression
suicide prediction cases had substantially longer follow-ups Scale significantly predicted ideation; there were too few cases of
(median 156 months; mean 181.35 months; s.d. = 156.35 months) the Hamilton Rating Scale for Depression predicting ideation to
than suicide attempt (median 36 months; mean 67.31 months; s. produce a reliable estimate. Although all three significantly pre-
d. 68.40 months) and ideation (median 24 months; mean 70.79 dicted attempts, none predicted suicide death (Table 2).
months; s.d. = 95.03 months). The subcategory, clinical features of depression, included pre-
dictors used to specify the nature of depression, such as age and
Effects on subsequent suicide ideation, attempt and type of onset (for example insidious) as well as the number, severity,
death duration and specific type (for example melancholic) of episodes.
Publication bias and overall prediction
Considered jointly, clinical features were only significant predictors
for attempts (wOR = 1.57, 95% CI 1.23–2.01) but not death; not
Overall prediction estimates reflect the pooled effect of all predictors enough cases were available in the prediction of ideation.
on the outcome of interest, regardless of category type. We also tested the effects of specific clinical features of depres-
sion. Number of episodes (wOR = 1.46, 95% CI 1.02–2.11) and
Ideation: OR analyses included 66 cases, generating a weighted course (wOR = 2.40, 95% CI 1.35–4.26) significantly increased the
mean OR (wOR) of 1.96 (95% CI 1.81–2.13). Publication bias was odds of attempt. Type of depressive episode was the only feature
evident across several indices (Table 1; Fig. 2). Between-study het- to significantly predict suicide death (wOR = 1.45, 95% CI 1.08–
erogeneity was high (I2 = 95.13%). Results were similar when com- 1.96). Cases were insufficient to examine finer-grained subcategor-
plete case dependence was assumed (wOR = 2.01, 95% CI 1.84– ies of course and episode type. However, most cases in the ‘course’
2.20). label reflect longer or more chronic illness. As such, comparisons
examine longer illness duration v. shorter illness duration. The
Attempt: A total of 166 prediction cases were included, yielding ‘episode type’ label includes a range of specifiers characterizing
a wOR of 1.63 (95% CI 1.55–1.72). Heterogeneity was high (I2 = the particular type (such as psychotic, atypical) of depressive
88.98%). Within models assuming complete dependence, the episode. Comparisons on this level examine the presence v.
wOR was 1.64 (95% CI 1.54–1.73). Publication bias was evident absence of an episode-type specifier. Although other features have
across multiple indices (Table 1; Fig. 2). been tested (for example age of onset, comorbid presentation),
too few cases existed in each category to produce reliable estimates.
Death: Analyses included 116 prediction cases, producing a The category, unspecified mood disorder diagnosis, included all
wOR of 1.33 (95% CI 1.18–1.49). Publication bias was detected prediction cases that used unspecified ‘mood disorder’ or ‘affective
across multiple indices (Table 2; Fig. 1). Heterogeneity was high disorder’ as predictors. Results indicated that unspecified mood
(I2 = 89.21%). Results were consistent within models assuming disorder diagnosis was only a significant predictor of death (wOR =
complete case dependence (wOR = 1.58, 95% CI 1.34–1.86). 1.64, 95% CI 1.11–2.42). Several cases examined the effects of
dysthymia; however, only attempt cases had a sufficient number of
Risk-factor category analyses cases to produce a reliable estimate, which was not significant.
To ensure reliability of estimates, more than three prediction cases Too few cases were available for any outcome for family history
were required for each risk-factor category analysis. See Table 2 for of depression.
results of all risk-factor category analyses.
Moderator analyses: overall prediction
Hopelessness: Hopelessness significantly predicted all out-
comes. The strongest effect was on ideation (wOR = 2.19, 95% CI Sample age
1.60– 3.00); estimates were weaker predicting attempt (wOR = Predicting ideation, adult samples produced statistically equivalent
1.95, 95% CI 1.59–2.39) and death (wOR = 1.98, 95% CI 1.46–2.69). effects (wOR = 2.15, 95% CI 1.89–2.46) relative to adolescent
samples (wOR = 1.95, 95% CI 1.66–2.30, χ2(1) = 0.84, P = 0.34);
Depression-related disorders and symptoms: A diagno- not enough cases were available to reliably report the effect on idea-
sis of major depressive disorder (MDD) was associated with signifi- tion for mixed samples (n = 3). Effects were statistically equivalent
cantly increased odds of ideation (wOR = 2.48, 95% CI 1.32–4.67), across age groups in the prediction of attempts (adults: wOR =

Table 1 Publication biasa

Fail-safe N Begg & Mazumdar rank Egger’s test of the Duval & Tweedie’s trim and fill
Classic Orwin’s correlation, τ (P) intercept, B0 (P) Missing cases Adjusted wOR (95% CI)
Suicidal ideation 1970 62 −0.12 (0.15) 3.22 (<0.001) 27 1.23 (1.13–1.34)
Suicide attempt 31675 157 −0.06 (0.27) 1.91 (<0.001) 45 1.33 (1.27–1.41)
Suicide death 4478 23 −0.13 (0.03) 1.02 (<0.005) 11 1.22 (1.08–1.38)
wOR, weighted mean odds ratio.
a. Classic and Orwin’s fail-safe N values represent the number of studies required to nullify the observed effects; Begg & Mazumdar rank correlation test computes the rank-order correlation
between effect estimates and standard error; Egger’s test of the intercept uses precision (i.e. the inverse of the standard error) to predict the standardised effect (i.e. effect size divided by the
standard error). The size of the effect is reflected in the slope and bias is reflected in the intercept (B0); missing cases under Duval & Tweedie’s trim and fill are the number of cases estimated
as missing below the mean.

4
 Depression and hopelessness as risk factors for suicide

(a)
0.0

0.5
Standard error

1.0

1.5

2.0
–3 –2 –1 0 1 2 3
Log odds ratio

(b)
0.0

0.5
Standard error

1.0

1.5

2.0
–3 –2 –1 0 1 2 3
Log odds ratio

(c)
0.0

0.5
Standard error

1.0

1.5

2.0
–5 –4 –3 –2 –1 0 1 2 3 4 5
Log odds ratio

Fig. 2 Funnel plots. (a) suicide ideation, (b) suicide attempt and (c) suicide death.
Open circles represent observed estimates; shaded circles represent imputed values estimated to be missing to the left of the mean (because of publication bias). Open diamond
indicates unadjusted weighted mean odds ratio; shaded diamond indicates weighted mean odds ratio adjusted for publication bias.

1.62, 95% CI 1.50–1.75); adolescents: wOR = 1.71, 95% CI 1.58– Follow-up length
1.86; mixed: wOR = 1.95, 95% CI 1.12–3.39, χ2(2) = 1.21, P = 0.55)
and death (adults: wOR = 1.30, 95% CI 1.14–1.48; mixed: wOR = There was no significant effect of follow-up length on the prediction
1.41, 95% CI 1.18–1.68), χ2(1) = 0.54, P = 0.46); there were too few of ideation (b = 0.0003, P = 0.79), attempt (b = 0.0003, P = 0.72) or
cases among adolescents predicting suicide death. death (b = −0.0005, P = 0.36).

5
Ribeiro et al

Table 2 Risk-factor category analyses across outcomes

Suicide ideation Suicide attempt Suicide death

Risk-factor categories n wOR 95% CI n wOR 95% CI n wOR 95% CI
Hopelessness 8 2.19 1.60–3.00 24 1.95 1.59–2.38 12 1.98 1.46–1.69
Depression-related disorders and symptoms
Major depressive disorder diagnosis 11 2.48 1.32–4.67 31 2.38 1.84–3.07 17 1.50 1.04–2.17
Depression symptoms 34 1.57 1.45–1.70 61 1.30 1.23–1.37 43 1.28 1.16–1.41
Self-report inventories 33 1.54 1.43–1.67 53 1.32 1.24–1.40 28 1.07 0.77–1.49
Beck Depression Inventory 5 1.93 1.50–2.48 17 1.15 1.06–1.25 8 1.04 0.63–1.71
Hamilton Rating Scale for Depression 1a – 12 1.53 1.17–1.99 2a – –
Center for Epidemiologic Studies Depression Scale 11 1.95 1.53–2.49 8 1.56 1.22–2.00 2a – –
Clinical features of depression 2a – – 29 1.57 1.23–2.01 28 1.00 0.99–1.01
Number of major depressive episodes 1a – – 11 1.46 1.02–2.11 7 0.83 0.60–1.15
Major depressive disorder type 1a – – 9 1.27 0.75–2.15 12 1.45 1.08–1.96
Course – – – 6 2.40 1.35–4.26 3a – –
Onset type – – – 3a – – 6 1.07 0.80–1.43
Dysthymia 2a – – 9 1.70 0.99–2.90 2a – –
Unspecified mood disorder 4 2.02 0.43–9.47 5 2.22 0.73–6.79 9 1.64 1.11–2.42

Family history of depression 2a – – 2a – – – – –

n, number of prediction cases; wOR, weighted mean odds ratio; –, indicates that information was not available.
a. Estimates are not reported for analyses involving three or fewer cases, as the small number of cases compromise the reliability of estimates; risk-factor categories with fewer than three
cases across all three outcomes are not listed in the table.

Sample severity Sample severity

Predicting ideation, effects were significantly stronger among general Predicting ideation, effects of depression-related factors were sig-
population samples (wOR = 2.60, 95% CI 2.21–3.05) than clinical nificantly stronger among general (wOR = 2.47, 95% CI 2.10–
(wOR = 1.31, 95% CI 1.17–1.47; χ2(1) = 45.96, P < 0.001) and self- 2.90) than clinical samples (wOR = 1.39, 95% CI 1.21–1.59, χ2(1) =
injurious samples (wOR = 1.12, 95% CI 1.00–1.25; χ2(1) = 69.19, 27.74, P < 0.001) and self-injurious samples (wOR = 1.61, 95% CI
P < 0.001). Prediction of ideation was significantly stronger among 0.72–3.62); χ2(2) = 28.53, P < 0.001). Effects were also significantly
clinical relative to self-injurious samples (χ2(1) = 3.85, P = 0.05). stronger among general population samples (wOR = 2.12, 95% CI
Predicting attempts, effects were significantly stronger among 1.91–2.34) in the prediction of attempts (clinical: wOR = 1.38, 95%
general (wOR = 2.00, 95% CI 1.82–2.20) than clinical (wOR = 1.52, CI 1.29–1.47; χ2(1) = 49.73, P < 0.001; self-injurious: wOR = 1.37,
95% CI 1.42–1.64, χ2(1) = 20.50, P < 0.001) and self-injurious 95% CI 1.03–1.81, χ2(1) = 8.25, P < 0.01). In the prediction of
samples (wOR = 1.49, 95% CI 1.15–1.94, χ2(1) = 4.28, P < 0.04); pre- suicide death, general samples demonstrated stronger effects (wOR
diction was statistically equivalent among self-injurious compared = 1.94, 95% CI 1.34–2.81) relative to clinical samples (wOR = 1.15,
with clinical samples (χ2(1) = 0.03, P = 0.88). Suicide death prediction 95% CI 1.01–1.32; χ2(1) = 6.83, P < 0.01) and statistically equivalent
was also significantly stronger among general (wOR = 2.01, 95% CI effects relative to self-injurious samples (wOR = 1.32, 95% CI 1.04–
1.40–2.87) compared with clinical (wOR = 1.17, 95% CI 1.03–1.34, 1.68, χ2(1) = 2.92, P = 0.09).
χ2(1) = 7.66, P < 0.01) samples, but statistically equivalent to self-
injurious samples (wOR = 1.43, 95% CI 1.12–1.82, χ2(1) = 2.42, P =
Moderator analyses: hopelessness effects
0.12); clinical and self-injurious samples were also statistically equiva-
Sample age
lent (χ2(1) = 1.76, P = 0.18).
Potential moderating effects of sample age on the effects of hope-
lessness on ideation and death prediction were not examined,
Moderator analyses: depression effects given an insufficient number of cases. In the prediction of ideation,
Sample age all cases were among adult samples; in the prediction of suicide
Effects of depression-related predictors were statistically equivalent death, all cases were among adult samples. Effects were statistically
across age groups in the prediction of ideation (adults: wOR = 2.25, equivalent among adult (wOR = 1.93, 95% CI 1.48–2.51) and ado-
95% CI 1.93–2.63; adolescents: 1.95, 95% CI 1.66–2.30; mixed: lescent (wOR = 1.92, 95% CI 1.48–2.48; χ2(1) = 0.001, P = 0.98)
insufficient cases; χ2(1) = 3.72, P = 0.05) and attempts (adults: samples in the prediction of attempts; too few cases existed using
wOR = 1.53, 95% CI 1.42–1.66; adolescents: wOR = 1.62, 95% CI mixed samples (n = 1).
1.50–1.76; mixed: wOR = 1.64, 95% CI 0.98–2.74; χ2(2) = 0.96, P =
0.62). Prediction was statistically equivalent among mixed Follow-up length
samples (wOR = 1.41, 95% CI 1.18–1.68) relative to adult samples There were no significant moderating effects of follow-up length
(wOR = 1.24, 95% CI 1.09–1.42; χ2(1) = 1.27, P = 0.26) in the predic- on the prediction of ideation (b = −0.006, P = 0.30), attempt (b =
tion of suicide death; there were too few cases among adolescent −0.002, P = 0.40) or death (b = −0.004, P = 0.30).
samples.

Sample severity
Follow-up length Moderating effects of sample severity for ideation were not con-
There was no significant effect of follow-up length on the pre- ducted as there were too few cases in each group. There were no
dictive power of depression-related factors across any outcome significant effects of sample severity in the prediction of suicide
(ideation: b = −0.0003, P = 0.54; attempt: b = 0.0007, P = 0.50; attempt (general: wOR = 1.91, 95% CI 1.34–2.73; clinical: wOR =
death: b = −0.0003, P = 0.55). 1.98, 95% CI 1.51–2.59; self-injurious: wOR = 2.35, 95% CI

6
 Depression and hopelessness as risk factors for suicide

1.02–5.40, χ2(2) = 0.20, P = 0.91) or death (general: too few cases First, most studies involved extremely long follow-ups. The
(n = 2); clinical: wOR = 1.69, 95% CI 1.07–2.66; self-injurious: average follow-up was 9 years, with very few studies focusing on
wOR = 2.38, , 95% CI 1.34–4.24, χ2(1) = 0.84, P = 0.36)). short-term prediction. This represents a critical gap in knowledge,
given existing clinical demands to assess risk over the course of
days or weeks. It is notable, though, that the effects of these
Discussion factors, although modest, do appear to be long lasting. As such,
these factors may have stronger effects over more clinically useful
Main findings (i.e. shorter) time frames.
Depression and hopelessness do confer risk for suicide ideation, Second, risk factors were typically measured as static, trait-like
attempt and death. At least when studied within the narrow meth- phenomena. As a result, virtually no studies examined how
odological constraints of the existing literature, overall prediction changes in these predictors could influence risk. Suicide risk is
estimates did not exceed wORs of 2.0 for any outcome, with the often conceptualised as transient and fluctuating, however.
weakest effects produced in the prediction of death by suicide. Assessing and studying these risk factors as state-like phenomena
Effects remained weak regardless of sample age or severity, or may be fruitful.
study follow-up length. Publication bias was evident across all out- Third, most studies examined the effects of risk factors in isola-
comes, and most pronounced for ideation and attempt prediction. tion. Advancing prediction accuracy will likely require simultaneous
Accounting for publication bias estimates further reduced effects, consideration of many risk factors and the complex relationships
suggesting the actual effects of depression and hopelessness are between those factors.33 Recent efforts applying machine learning
likely smaller than published research indicates. to prediction efforts support this approach.34,35 Thus, when consid-
Prediction of ideation, attempts and death varied slightly across ered in combination with many other factors in an optimal fashion,
risk-factor categories. The strongest predictors of suicide ideation depression and hopelessness may be very important predictors of
were hopelessness, BDI scores and an MDD diagnosis. For suicide suicidal behaviour.
attempts, an MDD diagnosis and course yielded the strongest
effects. Among the few factors that predicted suicide death, hope- Implications
lessness, unspecified mood disorder diagnosis and an MDD diagno- In closing, depression and hopelessness do appear to increase
sis were the most robust. suicide risk. Effects were weaker than expected; however, this may
Although some moderator effects emerged, results remained be an artefact of methodological constraints shared across the
largely unchanged. One consistent notable finding, however, was existing literature. Results from this investigation highlight a need
that effects appeared stronger among general relative to clinical to re-evaluate our current approaches to risk-detection strategies,
and self-injurious samples. We reason that this is likely a methodo- especially those that rely solely on the presence of depression or
logical artefact. Study designs that include clinical and self-injurious hopelessness. Recent machine learning efforts indicate that accurate
samples typically involve more stringent comparison groups; conse- prediction is possible; however, it will likely require consideration of
quently, these designs necessarily control for a host of risk factors complex combinations of many risk factors. The present work also
not usually accounted for in designs using general samples. highlights substantive gaps in knowledge. Research designs that
Selecting more homogenous samples (for example, clinical or self- examine the potential protean nature of these factors, especially in
injurious) can also result in a restricted range of the predictors the context of other sources of risk, to predict suicidal thoughts
and outcomes. This in turn can limit the ability to detect significant and behaviours in the short term will likely be particularly valuable.
associations. Of note, this pattern of findings echoes findings from We look forward to future work that advances beyond the methodo-
prior meta-analyses (for example Bentley et al31 and Ribeiro et al32). logical constraints of the existing literature in order to make mean-
To evaluate clinical significance, we considered the magnitude of ingful inroads in the prediction and prevention of suicidal thoughts
effects given the absolute risk of our outcomes. Suicidal thoughts and and behaviours.
behaviours are rare. In a given year, the prevalence of suicide deaths
in the USA is 0.00013. The strongest predictor of suicide death in this
Jessica D. Ribeiro, PhD, Xieyining Huang, BA, Department of Psychology, Florida
meta-analysis was hopelessness (wOR = 1.98). Doubling the odds of State University, Tallahassee, Florida; Kathryn R. Fox, MA, Department of Psychology,
suicide death only marginally improves our ability to detect risk of Harvard University, Cambridge, Massachusetts; Joseph C. Franklin, PhD, Department
of Psychology, Florida State University, Tallahassee, Florida, USA
death by suicide, given its extremely low prevalence. Moreover,
most clinicians are tasked with predicting risk over the course of Correspondence: Jessica D. Ribeiro, PhD, Department of Psychology, Florida State
University, 1107 W. Call St., Tallahassee, FL 32306-4301, USA. Email: [email protected]
days or weeks, making suicidal behaviours even rarer still.
Therefore, although hopelessness is strong relative to other risk First received 13 Nov 2017, final revision 1 Feb 2018, accepted 2 Feb 2018

factors, it is still weak in an absolute sense. This raises concerns

about relying solely on depression and hopelessness to detect risk.

Interpretation of our results and methodological

constraints Supplementary material
At first glance, our results appear to stand at odds with decades of
Supplementary material is available online at https://doi.org/10.1192/bjp.2018.27.
prominent suicide theories and clinical wisdom. We caution
against equating findings from this meta-analysis to the true
nature of these phenomena, however. Results of this meta-analysis References
may or may not reflect how these constructs actually relate to sui-
cidal thoughts and behaviours; instead, results reflect how these
1 Crosby AE, Han B, Ortega LAG, Parks SE, Gfroerer B. Suicidal thoughts and
constructs have been studied to date. The findings of this meta- behaviors among adults aged ≥18 years — United States, 2008–2009. MMWR
analysis also necessarily reflect any methodological constraints Surveill Summ 2011; 60: 1–22.
shared across the existing literature. We identified three major 2 World Health Organization. Suicide. WHO, 2015 (http://www.who.int/media-
methodological issues. centre/factsheets/fs398/en/).

7
Ribeiro et al

3 Centers for Disease Control and Prevention. Web-based Injury Statistics Query 21 Holma KM, Haukka J, Suominen K, Valtonen HM, Mantere O, Melartin TK, et al.
and Reporting System (WISQARS). CDC, no date (http://www.cdc.gov/ncipc/ Differences in incidence of suicide attempts between bipolar I and II disorders
wisqars). and major depressive disorder. Bipolar Disord 2014; 16: 652–61.
4 Kraemer HC, Kazdin AE, Offord DR, Kessler RC, Jensen PS, Kupfer DJ. Coming to 22 Niméus A, Alsén M, Träskman-Bendz L. The suicide assessment scale: an
terms with the terms of risk. Arch Gen Psychiatry 1997; 54: 337–43. instrument assessing suicide risk of suicide attempters. Eur Psychiatry 2000;
5 LeFevre M. Screening for suicide risk in adolescents, adults, and older adults in 15: 416–23.
primary care: US preventive services task force recommendation statement. 23 Ribeiro JD, Pease JL, Gutierrez PM, Silva C, Bernert RA, Rudd MD, et al. Sleep
Ann Intern Med 2014; 160: 719–26. problems outperform depression and hopelessness as cross-sectional and
6 Joiner T. Why People Die by Suicide. Harvard University Press, 2005. longitudinal predictors of suicidal ideation and behavior in young adults in the
military. J Affect Disord 2012; 136: 743–50.
7 Van Orden KA, Witte TK, Cukrowicz KC, Braithwaite SR, Selby EA, Joiner TE. The
interpersonal theory of suicide. Psychol Rev 2010; 117: 575–600. 24 Nock MK, Banaji MR. Prediction of suicide ideation and attempts among adoles-
cents using a brief performance-based test. J Consult Clin Psychol 2007; 75: 707–15.
8 Beck AT, Kovacs M, Weissman A. Hopelessness and suicidal behavior. An
overview. JAMA 1975; 234: 1146–9. 25 Keller F, Wolfersdorf M. Hopelessness and the tendency to commit suicide in the
course of depressive disorders. Cris J Cris Interv Suicide Prev 1993; 14: 173–7.
9 Abramson LY, Alloy LB, Hogan ME, Whitehouse WG, Gibb BE, Hankin BL, et al.
The hopelessness theory of suicidality. In Suicide Science: (eds T Joiner and MD 26 Keilp JG, Oquendo MA, Stanley BH, Burke AK, Cooper TB, Malone KM, et al.
Rudd): 17–32. Springer US, 2000. Future suicide attempt and responses to serotonergic challenge.
Neuropsychopharmacology 2010; 35: 1063–72.
10 American Psychiatric Association. Diagnostic and Statistical Manual of Mental
Disorders (DSM-5). APA, 2013. 27 Sher L, Carballo JJ, Grunebaum MF, Burke AK, Zalsman G, Huang YY, et al. A
prospective study of the association of cerebrospinal fluid monoamine
11 Nock MK, Hwang I, Sampson NA, Kessler RC. Mental disorders, comorbidity and metabolite levels with lethality of suicide attempts in patients with bipolar
suicidal behavior: results from the National Comorbidity Survey Replication. disorder. Bipolar Disord 2006; 8: 543–50.
Mol Psychiatry 2010; 15: 868–76.
28 Franklin JC, Ribeiro JD, Fox KR, Bentley KH, Kleiman EM, Huang X, et al. Risk
12 Conwell Y, Duberstein P, Cox C, Herrmann J, Forbes N, Caine E. Relationships of factors for suicidal thoughts and behaviors: a meta-analysis of 50 years of
age and axis I diagnoses in victims of completed suicide: a psychological aut- research. Psychol Bull 2016; 143: 187–232.
opsy study. Am J Psychiatry 1996; 153: 1001–8.
29 Thompson CG, Becker BJ. The impact of multiple endpoint dependency on Q
13 Harwood D, Hawton K, Hope T, Jacoby R. Psychiatric disorder and personality and I2 in meta-analysis. Res Synth Methods 2014; 5: 235–53.
factors associated with suicide in older people: a descriptive and case-control
study. Int J Geriatr Psychiatry 2001; 16: 155–65. 30 Scammacca N, Roberts G, Stuebing KK. Meta-analysis with complex research
designs: dealing with dependence from multiple measures and multiple group
14 Sareen J, Cox BJ, Afifi TO, de Graaf R, Asmundson GJG, ten Have M, et al. comparisons. Rev Educ Res 2014; 84: 328–64.
Anxiety disorders and risk for suicidal ideation and suicide attempts: a
population-based longitudinal study of adults. Arch Gen Psychiatry 2005; 62: 31 Bentley KH, Franklin JC, Ribeiro JD, Kleiman EM, Fox KR, Nock MK. Anxiety and
1249–57. its disorders as risk factors for suicidal thoughts and behaviors: a meta-analytic
review. Clin Psychol Rev 2016; 29: 30–46.
15 Miranda R, Gallagher M, Bauchner B, Vaysman R, Marroquín B.
Cognitive inflexibility as a prospective predictor of suicidal ideation 32 Ribeiro JD, Franklin JC, Fox KR, Bentley KH, Kleiman EM, Chang BP, et al. Self-
among young adults with a suicide attempt history. Depress Anxiety 2012; 29: injurious thoughts and behaviors as risk factors for future suicide ideation,
180–6. attempts, and death: a meta-analysis of longitudinal studies. Psychol Med
2016; 46: 225–36.
16 Klein DN, Schwartz JE, Rose S, Leader JB. Five-year course and outcome of
dysthymic disorder: a prospective, naturalistic follow-up study. Am J Psychiatry 33 Ribeiro JD, Franklin JC, Fox KR, Bentley KH, Kleiman EM, Chang BP, et al. Letter
2000; 157: 931–9. to the editor: suicide as a complex classification problem: machine learning and
related techniques can advance suicide prediction - a reply to Roaldset (2016).
17 O’Connor RC, Smyth R, Ferguson E, Ryan C, Williams JMG. Psychological pro- Psychol Med 2016; 46: 2009–10.
cesses and repeat suicidal behavior: a four-year prospective study. J Consult
Clin Psychol 2013; 81: 1137–43. 34 Kessler RC, Warner CH, Ivany C, Petukhova MV, Rose S, Bromet EJ, et al.
Predicting suicides after psychiatric hospitalization in US army soldiers: the
18 Kuo CJ, Tsai SY, Lo CH, Wang YP, Chen CC. Risk factors for completed suicide in Army Study to Assess Risk and Resilience in Service Members (Army STARRS).
schizophrenia. J Clin Psychiatry 2005; 66: 579–85. JAMA Psychiatry 2015; 72: 49.
19 Beck AT, Steer RA, Kovacs M, Garrison B. Hopelessness and eventual suicide: a 35 Walsh CG, Ribeiro JD, Franklin JC. Predicting risk of suicide attempts over time
10-year prospective study of patients hospitalized with suicidal ideation. Am J through machine learning. Clin Psychol Sci 2017; 5: 457–69.
Psychiatry 1985; 142: 559–63.
20 Juon HS, Ensminger ME. Childhood, adolescent, and young adult predictors of
suicidal behaviors: a prospective study of African Americans. J Child Psychol
Psychiatry 1997; 38: 553–63.