Severity and Variability of Depression Symptoms Predicting Suicide Attempt in High-Risk Individuals

Nadine M Melhem; Giovanna Porta; Maria A Oquendo; Jamie Zelazny; John G Keilp; Satish Iyengar; Ainsley Burke; Boris Birmaher; Barbara Stanley; J John Mann; David A Brent

doi:10.1001/jamapsychiatry.2018.4513

. 2019 Feb 27;76(6):603–613. doi: 10.1001/jamapsychiatry.2018.4513

Severity and Variability of Depression Symptoms Predicting Suicide Attempt in High-Risk Individuals

Nadine M Melhem ^1,^✉, Giovanna Porta ², Maria A Oquendo ³, Jamie Zelazny ¹, John G Keilp ⁴, Satish Iyengar ⁵, Ainsley Burke ⁴, Boris Birmaher ¹, Barbara Stanley ⁴, J John Mann ⁴, David A Brent ¹

¹Department of Psychiatry, University of Pittsburgh School of Medicine, Pittsburgh, Pennsylvania

²University of Pittsburgh Medical Center, Pittsburgh, Pennsylvania

³Department of Psychiatry, Perelman School of Medicine, University of Pennsylvania, Philadelphia

⁴Department of Psychiatry, Columbia University, New York, New York

⁵Department of Statistics, University of Pittsburgh, Pittsburgh, Pennsylvania

Accepted for Publication: November 25, 2018.

^✉

Corresponding Author: Nadine M. Melhem, PhD, Department of Psychiatry, University of Pittsburgh School of Medicine, 3811 O’Hara St, Pittsburgh, PA 15213 (melhemnm@upmc.edu).

Published Online: February 27, 2019. doi:10.1001/jamapsychiatry.2018.4513

Author Contributions: Drs Melhem and Brent had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Melhem, Keilp, Burke, Birmaher, Stanley, Mann, Brent.

Acquisition, analysis, or interpretation of data: Melhem, Porta, Oquendo, Zelazny, Keilp, Iyengar, Burke, Birmaher, Stanley, Mann, Brent.

Drafting of the manuscript: Melhem.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Melhem, Porta, Iyengar, Brent.

Obtained funding: Melhem, Keilp, Mann, Brent, Oquendo.

Administrative, technical, or material support: Melhem, Oquendo, Keilp, Mann, Brent.

Supervision: Melhem, Oquendo, Keilp, Mann, Brent.

Conflict of Interest Disclosures: .Dr Melhem reported receiving research support from the National Institute of Mental Health (NIMH), Brain and Behavior Research Foundation, and the American Foundation for Suicide Prevention (AFSP). Dr Oquendo reported receiving royalties from Research Foundation for Mental Hygiene for the commercial use of the Columbia-Suicide Severity Rating Scale, and owning stock in Bristol-Myers Squibb. Drs Burke, Stanley, and Mann reported receiving royalties from Research Foundation for Mental Hygiene for the commercial use of the Columbia-Suicide Severity Rating Scale. Dr Stanley also reported receiving research support from NIMH and AFSP. Dr Keilp reported receiving support from AFSP and owning stocks in Pfizer and Zoetis. Dr Birmaher reported receiving research support from the National Institute of Mental Health (NIMH) and royalties for publications from Random House Inc, Lippincott Williams & Wilkins, and UpToDate. Dr Iyengar reported receiving research support from NIMH. Dr Brent receives research support from NIMH, AFSP, and the Once Upon a Time Foundation, honoraria from the Klingenstein Third Generation Foundation royalties for scientific board membership and grant review, from Guilford Press, royalties from the electronic self-rated version of the C-SSRS from ERT, Inc., royalties from performing duties as an UpToDate Psychiatry Section Editor, payment for serving as an Associate Editor for Psychological Medicine, and consulting fees from Healthwise. No other disclosures were reported.

Funding/Support: This study was supported by R01 grants MH056612 (Dr Brent) and MH056390 (Dr Mann) and K01 grant MH077930 (Dr Melhem) from the NIMH.

Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Contributions: We would like to acknowledge the families for their participation in this study.

^✉

Corresponding author.

PMCID: PMC6551844 NIHMSID: NIHMS1020382 PMID: 30810713

Key Points

Question

What are the most important clinical predictors of suicide attempt?

Findings

In this longitudinal study of 663 offspring of parents with mood disorders, the trajectory of depression symptoms showing the highest mean scores and variability over time predicted suicide attempt above and beyond psychiatric diagnoses. Additional predictors were younger age (≤30 years), mood disorders, childhood abuse, and personal and parental history of suicide attempt.

Meaning

Predictors were identified that clinicians already assess during routine psychiatric evaluation; clinicians should especially monitor and treat depression symptoms to reduce the risk for suicidal behavior.

Abstract

Importance

Predicting suicidal behavior continues to be among the most challenging tasks in psychiatry.

Objectives

To examine the trajectories of clinical predictors of suicide attempt (specifically, depression symptoms, hopelessness, impulsivity, aggression, impulsive aggression, and irritability) for their ability to predict suicide attempt and to compute a risk score for suicide attempts.

Design, Setting, and Participants

This is a longitudinal study of the offspring of parents (or probands) with mood disorders who were recruited from inpatient units at Western Psychiatric Institute and Clinic (Pittsburgh) and New York State Psychiatric Institute. Participants were recruited from July 15, 1997, to September 6, 2005, and were followed up through January 21, 2014. Probands and offspring (n = 663) were interviewed at baseline and at yearly follow-ups for 12 years. Lifetime and current psychiatric disorders were assessed, and self-reported questionnaires were administered. Model evaluation used 10-fold cross-validation, which split the entire data set into 10 equal parts, fit the model to 90% of the data (training set), and assessed it on the remaining 10% (test set) and repeated that process 10 times. Preliminary analyses were performed from July 20, 2015, to October 5, 2016. Additional analyses were conducted from July 26, 2017, to July 24, 2018.

Main Outcomes and Measures

The broad definition of suicide attempt included actual, interrupted, and aborted attempts as well as suicidal ideation that prompted emergency referrals during the study. The narrow definition referred to actual attempt only.

Results

The sample of offspring (n = 663) was almost equally distributed by sex (316 female [47.7%]) and had a mean (SD) age of 23.8 (8.5) years at the time of censored observations. Among the 663 offspring, 71 (10.7%) had suicide attempts over the course of the study. The trajectory of depression symptoms with the highest mean scores and variability over time was the only trajectory to predict suicide attempt (odds ratio [OR], 4.72; 95% CI, 1.47-15.21; P = .01). In addition, we identified the following predictors: younger age (OR, 0.82; 95% CI, 0.74-0.90; P < .001), lifetime history of unipolar disorder (OR, 4.71; 95% CI, 1.63-13.58; P = .004), lifetime history of bipolar disorder (OR, 3.4; 95% CI, 0.96-12.04; P = .06), history of childhood abuse (OR, 2.98; 95% CI, 1.40-6.38; P = .01), and proband actual attempt (OR, 2.24; 95% CI, 1.06-4.75; P = .04). Endorsing a score of 3 or higher on the risk score tool resulted in high sensitivity (87.3%) and moderate specificity (63%; area under the curve = 0.80).

Conclusions and Relevance

The specific predictors of suicide attempt identified are those that clinicians already assess during routine psychiatric evaluations; monitoring and treating depression symptoms to reduce their severity and fluctuation may attenuate the risk for suicidal behavior.

This study evaluates and follows up the children of parents with psychiatric disorders to assess their risks and likelihood for suicidal ideation and suicide attempts.

Introduction

Suicide is the second leading cause of death in the United States among people aged 15 to 34 years.¹ History of attempts, mood disorders, and substance use disorders are the most important clinical predictors of suicidal behavior.^{2,3,4,5,6,7,8,9} Additional predictors include family history of suicidal behavior^8,10; impulsive aggression^8,10,11; lethality of method¹²; insomnia^12,13; hopelessness, anxiety, and agitation^{7,14,15,16,17,18}; and childhood adversity and negative life experiences.^{19,20,21,22,23,24,25} In a meta-analysis of 50 years of research, the prediction of suicidal ideation, attempt, and suicide deaths with these predictors was only slightly better than chance.²⁶ Machine learning approaches have been applied to estimating the risk for suicidal behavior using electronic health records^27,28,29,30 with improved performance.^28,30,31 Predictors included lifetime psychiatric diagnoses and substance abuse.^28,29,30,31

Psychiatric diagnoses are well-established indicators of suicidal behavior. However, diagnoses and stable or trait-like predictors are of limited value because the risk of suicidal behavior varies during the course of psychiatric illness. Thus, it is important to identify symptoms that vary over time. We examined the trajectories of impulsivity, aggression, impulsive aggression, depression symptoms, irritability, and hopelessness in a longitudinal study of offspring of parents with mood disorders. We examined whether changes in these measures over time predict attempt and time to onset of attempt above and beyond psychiatric diagnoses and other predictors. In addition, we computed a risk score based on our models and assessed its performance.

Methods

The study protocol was approved by the institutional review boards at Western Psychiatric Institute and Clinic (Pittsburgh, Pennsylvania) and New York State Psychiatric Institute (New York). Written informed consent was obtained from adults, and written parental consent and assent were obtained for those younger than 18 years.

Sample

The sample consisted of 663 offspring of 318 parents with mood disorders from the larger sample of 711 offspring of 337 parents (eFigure 1 in the Supplement). Parents are referred hereafter as probands. Probands were recruited from inpatient units at Western Psychiatric Institute and Clinic and New York State Psychiatric Institute. All probands had a history of mood disorder, and 180 (56.6%) of them had a lifetime history of an actual suicide attempt. We excluded 48 offspring who were lost to follow-up after the baseline assessment. We compared them with the remaining 663 offspring and found them substantially different in some demographic and clinical characteristics, with the excluded offspring having higher rates of mood and psychotic disorders (eTable 1 in the Supplement). Participants were recruited from July 15, 1997, to September 6, 2005, and were followed through January 21, 2014.

We used a broad and a narrow definition of suicide attempt in offspring. The broad definition included an actual attempt or a suicide-related behavior (eg, interrupted, aborted, and ambiguous).^32,33,34 We also included suicidal ideation that prompted emergency referral during the study, consistent with treatment studies.^8,35 Our safety procedures included emergency referrals for those with suicidal ideation with intent and a plan at the time of assessment and thus may have prevented suicidal behavior. Here, we refer to the broad definition as suicide attempt and the narrow definition, including actual attempts only, as actual attempt.

Assessment

Probands and offspring were interviewed at baseline and yearly follow-ups for 12 years. We assessed lifetime and current psychiatric disorders at baseline as well as since the last assessment and at current follow-ups using the Schedule for Affective Disorders and Schizophrenia for School-Age Children—Present and Lifetime Version³⁶ and the Structured Clinical Interview for DSM-IV.³⁷ Similarly, suicidal behavior was assessed using the Columbia Suicide History Form and the Medical Lethality Rating Scale.³⁸ Self-reported questionnaires were administered to assess the current severity of symptoms: the Beck Depression Inventory³⁹ for those older than 18 years and the Children’s Depression Inventory⁴⁰ for those younger than 18 years; the Beck Hopelessness Scale⁴¹ and the Hopelessness Scale for Children⁴²; the Barratt Impulsiveness Scale⁴³ in adults and the subscales of emotionality, activity, sociability, and impulsivity in those younger than 18 years⁴⁴; and the Buss-Durkee Hostility Inventory⁴⁵ and its downward extension, the Children’s Hostility Inventory,⁴⁶ to measure impulsive aggression. The Brown Goodwin Aggression Scale⁴⁷ was used to assess lifetime history of aggressive behaviors at baseline, and follow-up assessments covered the time frame since the last assessment. We measured irritability using the irritability items from the depression scales. History of childhood abuse was assessed with the Childhood Trauma Questionnaire⁴⁸ and an adapted instrument from the Abuse Dimensions Inventory.⁴⁹ Standardized scores were computed when different measures were used for those younger or older than 18 years.

Statistical Analysis

Preliminary analyses were performed from July 20, 2015, to October 5, 2016. Additional analyses were conducted from July 26, 2017, to July 24, 2018. To identify the trajectories of symptoms for depression, hopelessness, irritability, impulsivity, aggression, and impulsive aggression, we used the lcmm package in R (R Foundation for Statistical Computing) to fit mixed models using latent classes. Observations were censored at the time point prior to the first suicide attempt for offspring with an event during the course of the study and at the time point of last observation for those without attempt. We used the hlme function in R to fit latent class linear mixed models, which take into account within-subject variability. We used the unstructured variance-covariance matrix and allowed the variance-covariance of the random effects to be class specific, which does not impose any constraints on the model so as to better estimate variability patterns. We used the Bayesian Information Criteria to identify the best-fitting model. We examined the concordance between the trajectories using percent agreement and Cohen κ coefficient.

We compared the demographic and clinical characteristics of those with and without attempt using χ² tests; Fisher exact test; and unpaired, 2-tailed t tests. Similarly, we compared the resulting trajectories on these characteristics. We applied a Bonferroni correction for multiple comparisons for the 6 measures (2-sided α < .01 or α < .05/6 indicates statistical significance): depression symptoms, hopelessness, irritability, impulsivity, aggression, and impulsive aggression. We then compared offspring with and without attempt on the resulting trajectories for each of the 6 measures, and we examined with logistic regression whether these trajectories predicted attempt.

Regression models were conducted with the broad and narrow definitions of attempt. For all analyses, we used 2 sets of regression models. Model 1 controlled for age, sex, race/ethnicity, site, and income. We examined the trajectories of income to account for changes over time (eFigure 2 in the Supplement). We also controlled for proband lifetime history of actual attempt because the study design included recruiting probands, half of whom had such a history, and for offspring lifetime history of attempt at baseline. Model 2 controlled for all variables in model 1 and for offspring lifetime history of the most common psychiatric disorders in our sample up to the time of attempt, history of childhood abuse, proband history of bipolar disorders, psychiatric treatment at the time point prior to attempt, and cluster B personality disorders in offspring 18 years of age or older (n = 507). We then included all symptom trajectories into 1 regression model.

Similarly, we used Cox proportional hazards regression models to examine time to onset of attempt, with psychiatric disorders and psychiatric treatment in model 2 as time-varying covariates. Because of multiple siblings per family, we estimated the variance-covariance matrix with correlated observations and computed clustered robust SEs. Regression models were also used to predict incident attempt (broad definition). We were not able to use these models for incident actual attempt given their small number (n = 19). For model evaluation, we used 10-fold cross-validation in which we split the entire data set into 10 equal parts, fit the model to 90% of the data (training set), and assessed it on the remaining 10% (test set) and repeated that process 10 times.

We applied the multiple imputation by chained equations technique on covariates with missing data only in Stata, ice command (StataCorp LLC).^50,51 There were no missingness or missing data on 1 variable for 75% of the sample, 12% had missing data on 2 variables, and 13% had missing data on 3 or more variables. Similar results were obtained using the original and imputed data sets; thus, we report the results from the imputed data set. The final models on the nonimputed data set are reported in eTable 16 and eTable 18 in the Supplement. We also used machine learning regression approaches—LASSO (least absolute selection and shrinkage operator)—by using the glmnet package in R⁵² to determine a subset of predictors with the strongest effects. We computed a risk score based on prediction models and examined area under the curve, sensitivity, and specificity; positive predictive value; and negative predictive value.

Results

The sample of offspring (n = 663) was almost equally distributed by sex (316 female [47.7%]) and had a mean (SD) age of 23.8 (8.5) years at the time of censored observations (Table 1). Of the 663 offspring, 455 of 647 (70.3%) were white individuals, and the median (range) follow-up was 8.1 (1-15.4) years. Compared with offspring who completed follow-up for at least 8 years, those with fewer than 8 years of follow-up or who were lost to follow-up were younger (mean [SD] age, 22.1 [9.9] years vs 26.2 [5.8] years; t = 6.70; df = 631.02; P < .001) and less likely to have an anxiety disorder (21.3% vs 31.2%; χ²₁= 8.36; P = .004).

Table 1. Demographic and Clinical Characteristics and Trajectories of Symptoms for Offspring .

Variable	No. (%)			Test	df	P Value	Cohen d
Variable	Total Sample	Without Suicide Attempt	With Suicide Attempt	Test	df	P Value	Cohen d
Total^a	663	592	71	NA	NA	NA	NA
Demographics and clinical characteristics
Age, mean (SD), y	23.8 (8.5)	24.5 (8.7)	18.5 (5.8)	t: 7.63	111.43	<.001	0.70
Female sex	316 (47.7)	280 (47.3)	36 (50.7)	χ²₁: 0.30	1	.59	0.07
White race/ethnicity	455/647 (70.3)	411/576 (71.4)	44 (62.0)	χ²₁: 2.67	1	.10	−0.21
Pittsburgh site	443 (66.8)	391 (66.0)	52 (73.2)	χ²₁: 1.48	1	.22	0.15
Class 2 income	435/660 (65.9)	398/590 (67.5)	37/70 (52.9)	χ²₁: 5.94	1	.02	−0.31
Income, mean (SD)^b	5.3 (3.0)	5.5 (3.0)	4.3 (2.8)	t: 2.99	613	.003	0.39
Lifetime history of psychiatric disorders
Mood disorders^c	360 (54.3)	301 (50.8)	59 (83.1)	χ²₁: 26.58	1	<.001	0.66
No mood disorder	303 (47.5)	291 (49.2)	12 (16.9)	χ²₁: 26.58	1	<.001	0.66
Unipolar disorder	301 (45.4)	253 (42.7)	48 (67.6)	χ²₁: 26.95	2	<.001	0.63
Bipolar disorder	59 (8.9)	48 (8.1)	11 (15.5)	χ²₁: 26.95	2	<.001	0.63
No. of depression episodes, mean (SD)	1.1 (2.2)	1.1 (2.2)	1.6 (1.9)	t: −2.05	661	.04	0.26
Duration of depression episodes, mean (SD), mo	19.4 (27.2)	17.5 (26.0)	28.2 (31.2)	t: –2.28	65.42	.03	0.40
Anxiety	284 (42.8)	248 (41.9)	36 (50.7)	χ²₁: 2.01	1	.16	0.18
Disorders
Psychotic	7 (1.1)	5 (0.8)	2 (2.8)	FET	NA	.17	0.19
Alcohol and substance use	220 (33.2)	196 (33.1)	24 (33.8)	χ²₁: 0.01	1	.91	0.01
Posttraumatic stress	87 (13.1)	74 (12.5)	13 (18.3)	χ²₁: 1.88	1	.17	0.17
Attention-deficit/hyperactivity	115/630 (18.3)	98/569 (17.2)	17/61 (27.9)	χ²₁: 4.18	1	.04	0.28
Cluster B personality^d	55/502 (11.0)	45/463 (9.7)	10/39 (25.6)	FET	NA	.01	0.52
Proband
Bipolar	217/651 (33.3)	185/581 (31.8)	32/70 (45.7)	χ²₁: 5.41	1	.02	0.30
Attempt	391 (59.0)	339 (57.3)	52 (73.2)	χ²₁: 6.69	1	.01	0.33
History
Suicide attempt	58 (8.8)	38 (6.4)	20 (28.2)	χ²₁: 37.57	1	<.001	0.79
Childhood abuse	181/556 (32.6)	142/487 (29.2)	39/69 (56.5)	χ²₁: 20.61	1	<.001	0.60
Psychiatric treatment at time point prior to censoring
Inpatient	18/633 (2.8)	11/563 (2.0)	7/70 (10.0)	FET	NA	.002	0.49
Outpatient	53/144 (36.8)	33/111 (29.7)	20/33 (60.6)	χ²₁: 10.43	1	.001	0.66
Psychotherapy	532/607 (87.6)	482/550 (87.6)	50/57 (87.7)	χ²₁: 0.0003	1	.99	0.003
Antidepressants/mood stabilizers	97 (14.6)	80 (13.5)	17 (23.9)	χ²₁: 5.52	1	.02	0.30
Other psychotropic medications	69 (10.4)	62 (10.5)	7 (9.9)	χ²₁: 0.03	1	.87	−0.02
Nonpsychotropic medications	235 (35.4)	213 (36.0)	22 (31.0)	χ²₁: 0.69	1	.41	−0.10
Trajectories of symptoms
Depression symptoms class
2	174/647 (26.9)	151/577 (26.2)	23/70 (32.9)	FET	NA	<.001	0.72
3	43/647 (6.6)	27/577 (4.7)	16/70 (22.9)	FET	NA	<.001	0.72
Class 2
Hopelessness	154 (25.4)	140 (24.1)	25 (35.7)	χ²₁: 4.42	1	.04	0.27
Impulsivity	151 (23.5)	125 (21.7)	26 (38.8)	χ²₁: 9.77	1	.002	0.41
Aggression	112/659 (17.0)	92/590 (15.6)	20/69 (29.0)	χ²₁: 7.85	1	.01	0.36
Impulsive aggression	349/637 (54.8)	297/569 (52.2)	52/68 (76.5)	χ²₁: 14.45	1	<.001	0.49
Irritability	159/649 (24.5)	131/578 (22.7)	28 (39.4)	χ²₁: 9.62	1	.002	0.39

Open in a new tab

Abbreviations: FET, Fisher exact test; NA, not applicable.

^{^a}

The denominator used to calculate the percentages may be different for different items because of missing data. Denominators that vary from column heading are included with number.

^{^b}

1 = <US $10 000; 2 = $10 000 to $14 999; 3 = $15 000 to $19 999; 4 = $20 000 to $29 999; 5 = $30 000 to $39 999; 6 = $40 000 to $49 999; 7 = $50 000 to $59 999; 8 = $60 000 to $69 999; and 9 = ≥$70 000.

^{^c}

Combining unipolar and bipolar disorders.

^{^d}

Only offspring aged ≥18 years included (n = 507).

Among the 663 offspring, 71 (10.7%) had suicide attempts over the course of the study (eFigure 1 in the Supplement), 51 of which were first-time attempts for an incidence rate of 8.4%. Among attempters, the mean (SD) number of suicide attempts was 1.2 (0.6) and of actual attempts was 1.3 (0.7). The mean (SD) lethality (for actual attempts only) was 1.7 (2), which corresponded to physical damage for which medical attention was needed. The median (range) time from the last assessment point to suicide attempt was 45 (1-126) weeks (30 [4-126] weeks for actual attempt).

Trajectory Analyses

We identified a 2-class model for hopelessness (mean [SD] score, 0.99 [0.78] vs –0.35 [0.34]; t_185.9 = –21.5; P < .001; Cohen d = 2.74), impulsivity (1.17 [0.60] vs –0.39 [0.57]; t₆₄₁ = –29; P < .001; Cohen d = 2.70), aggression (1.13 [0.71] vs –0.26 [0.32]; t_120.1 = –20.3; P < .001; Cohen d = 3.4), and irritability (0.92 [0.60] vs –0.29 [0.36]; t_195.8 = –24.1; P < .001; Cohen d = 2.83], in which class 2 on these measures consistently showed higher mean scores and variability compared with class 1 (Figure 1; eTable 2 in the Supplement). We also identified a 2-class model for impulsive aggression, in which class 2 showed higher mean scores compared with class 1 (0.60 [0.62] vs –0.79 [0.41]; t₆₀₇ = –33.8; P < .001; Cohen d = 2.60), although the variability patterns between the 2 classes were similar. For depression symptoms, we identified a 3-class model, in which class 3 showed the highest mean scores and variability (2.08 [0.75] vs 0.59 [0.42] vs –0.44 [0.28] (class 1); F_2,644 = 1216.7; P = <.001; Cohen d class 3 vs 2 = 2.97; Cohen d class 3 vs 1 = 7.31; Cohen d class 2 vs 1 = 3.17). eTable 3 in the Supplement shows the concordance between trajectories across the different measures. Depression and hopelessness showed the highest agreement (κ = 0.42; 74.6% agreement). eTables 4 to 9 in the Supplement show the demographic and clinical characteristics of the trajectories.

Figure 1. — A, Depression symptoms class 1 (n = 430), class 2 (n = 174), and class 3 (n = 43). B, Hopelessness class 1 (n = 485) and class 2 (n = 165). C, Impulsivity class 1 (n = 492) and class 2 (n = 151). D, Aggression class 1 (n = 547) and class 2 (n = 112). E, Impulsive aggression class 1 (n = 288) and class 2 (n = 349). F, Irritability class 1 (n = 490) and class 2 (n = 159). Only 34 offspring had a follow-up duration beyond 12 years. Data from these follow-ups were truncated and removed from the analyses. During the course of this longitudinal study, the impulsivity and impulsive aggression questionnaires were eliminated at year 8 to reduce the assessment battery.

Characteristics of Offspring With Suicide Attempt

Table 1 compares the offspring with and without suicide attempt. In addition to some clinical characteristics, we also found those with attempt to be more likely than those without attempt to belong to class 3 depression symptoms with higher mean and variability (16 [22.9%] vs 27 [4.7%]; P < .001; Cohen d = 0.72) (Table 1). They were also more likely to belong to class 2 for impulsivity (26 [38.8%] vs 125 [21.7%]; χ²₁ = 9.77; P = .002; Cohen d = 0.41), aggression (20 [29.0%] vs 92 [15.6%]; χ²₁ = 7.85; P = .01; Cohen d = 0.36), impulsive aggression (52 [76.5%] vs 297 [52.2%]; χ²₁ = 14.45; P < .001; Cohen d = 0.49), and irritability (28 [39.4%] vs 131 [22.7%]; χ²₁ = 9.62; P = .002; Cohen d = 0.39).

Individual Trajectories as Predictors of Suicide Attempt

Class 3 depression symptoms with higher mean and variability was the only statistically significant trajectory that predicted increased risk for suicide attempt in offspring (model 1 odds ratio [OR], 6.53 [95% CI, 2.53-16.87; t = 3.88; P < .001]; model 2 OR, 3.39 [95% CI, 1.32-8.66; t = 2.55; P = .01) (eTable 10, models 1-2 in the Supplement). Similar results were obtained when looking at the narrow definition of actual attempt (eTable 11, model 1 in the Supplement), incident attempt (eTable 12, model 1 in the Supplement), and time to onset of events (eTables 13-15, model 1 in the Supplement). However, none of the trajectories predicted time to onset of any of the events when controlling for psychiatric disorders and treatments as time-varying covariates (eTables 13-15, model 2 in the Supplement).

Combined Trajectories as Predictors of Suicidal Behavior

When we included the trajectories for all measures and controlled for demographics and proband history of actual attempt (Table 2, model 1; eTable 16, model 1 in the Supplement), we found that class 3 depression symptoms was associated with an almost 8-fold increased risk for suicide attempt (OR, 7.69; 95% CI, 2.37-24.90; t = 3.40; P = .001) and was the only statistically significant trajectory to predict attempt.

Table 2. Prediction of Suicide Attempt (Broad Definition) Combining Symptom Trajectories^a.

Variable	OR (95% CI)	t Test	P Value
Model 1
Depression symptoms class^b
2	1.51 (0.67-3.39)	1.00	.32
3	7.69 (2.37-24.90)	3.40	.001
Class 2^c
Hopelessness	0.61 (0.25-1.51)	−1.07	.29
Impulsivity	0.89 (0.42-1.91)	−0.29	.77
Aggression	1.75 (0.80-3.84)	1.39	.16
Impulsive aggression	1.38 (0.65-2.94)	0.84	.40
Irritability	1.05 (0.51-2.17)	0.14	.89
Age	0.87 (0.82-0.92)	−5.00	<.001
Sex, female vs male	1.04 (0.54-2.00)	0.12	.90
Race/ethnicity, white vs non-white	1.26 (0.67-2.39)	0.72	.47
Site, Pittsburgh vs New York	0.97 (0.47-2.01)	−0.09	.93
Income, class 2^c	0.53 (0.27-1.06)	−1.80	.07
History of suicide attempt	3.26 (1.50-7.11)	2.98	.003
Proband suicide attempt	1.88 (0.99-3.56)	1.94	.05
Model 2
Depression symptoms class^b
2	0.78 (0.33-1.86)	−0.56	.57
3	4.72 (1.47-15.21)	2.60	.01
Class 2^c
Hopelessness	0.77 (0.31-1.95)	−0.54	.59
Impulsivity	0.85 (0.35-2.04)	−0.36	.71
Aggression	1.23 (0.55-2.78)	0.50	.62
Impulsive aggression	1.24 (0.52-2.94)	0.48	.63
Irritability	0.81 (0.37-1.78)	−0.51	.61
Age	0.82 (0.74-0.90)	−4.11	<.001
Sex, female vs male	0.74 (0.37-1.48)	−0.85	.39
Race/ethnicity, white vs non-white	1.39 (0.68-2.84)	0.89	.37
Site, Pittsburgh vs New York	1.19 (0.52-2.72)	0.41	.68
Income, class 2^c	0.48 (0.24-0.95)	−2.11	.04
Disorder
Unipolar	4.71 (1.63-13.58)	2.87	.004
Bipolar	3.40 (0.96-12.04)	1.90	.06
Anxiety	0.93 (0.38-2.25)	−0.17	.87
Alcohol and substance use	1.73 (0.76-3.92)	1.31	.19
Posttraumatic stress disorder	1.20 (0.38-3.73)	0.31	.76
Attention-deficit/hyperactivity disorder	0.56 (0.21-1.49)	−1.17	.24
History
Suicide attempt	2.26 (0.98-5.23)	1.91	.06
Childhood abuse	2.98 (1.40-6.38)	2.82	.01
Proband
Actual attempt	2.24 (1.06-4.75)	2.11	.04
Bipolar disorder	1.20 (0.63-2.30)	0.55	.59
Inpatient	1.83 (0.47-7.20)	0.87	.38
Outpatient	1.60 (0.40-6.40)	0.68	.50
Psychotherapy	0.66 (0.19-2.28)	−0.66	.51
Antidepressant/mood stabilizer	1.91 (0.65-5.66)	1.18	.24
Other psychotropic medication	0.32 (0.08-1.22)	−1.67	.09

Open in a new tab

Abbreviation: OR, odds ratio.

^{^a}

The broad definition of suicide attempt included an actual attempt or a suicide-related behavior (interrupted, aborted, or ambiguous or suicidal ideation that prompted emergency referral during the study).

^{^b}

Dummy-coded variable with class 1 as the reference category.

^{^c}

Class 2 compared with class 1. Model 1 pseudo R² = 0.225; % correctly classified: 91.4% (full model) and 90.4% (test model). Model 2 pseudo R² = 0.326; % correctly classified: 92.4% (full model) and 90% (test model).

Class 3 depression symptoms remained the only statistically significant trajectory to predict suicide attempt when controlling for additional clinical characteristics (OR, 4.72; 95% CI, 1.47-15.21; t = 2.60; P = .01) (Table 2, model 2; eTable 16, model 2 in the Supplement). When including cluster B personality disorders in models, including offspring aged 18 years or older, we obtained similar results (eTable 17 in the Supplement). Similar results were obtained when using the narrow definition (Table 3; eTable 18 in the Supplement), incident attempt (eTable 19 in the Supplement), time to onset of suicide attempt (Figure 2; eTable 20 in the Supplement), and actual and incident attempts (eTables 21-22 in the Supplement). However, some of the predictors were no longer statistically significant, and some models were not stable given the reduced sample size for the narrow and incident attempt definitions. Using LASSO (eTable 23 in the Supplement), we found the following predictors in addition to class 3 depression symptoms: younger age, a lifetime history of unipolar and bipolar disorders, history of childhood abuse, and proband actual attempt.

Table 3. Prediction of Actual Attempt (Narrow Definition) Combining Symptom Trajectories^a.

Variable	OR (95% CI)	t Test	P Value
Model 1
Depression symptoms class^b
2	0.85 (0.26-2.72)	−0.28	.78
3	8.22 (1.71-39.57)	2.63	.01
Class 2^c
Hopelessness	0.48 (0.11-2.02)	−1.00	.32
Impulsivity	0.86 (0.26-2.78)	−0.26	.80
Aggression	1.19 (0.35-4.05)	0.28	.78
Impulsive aggression	2.38 (0.82-6.91)	1.60	.11
Irritability	0.62 (0.17-2.28)	−0.71	.48
Age	0.87 (0.79-0.95)	−3.01	.003
Sex, female vs male	1.44 (0.54-3.86)	0.73	.47
Race/ethnicity, white vs non-white	1.35 (0.55-3.36)	0.65	.51
Site, Pittsburgh vs New York	0.52 (0.18-1.50)	−1.21	.23
Income, class 2^c	0.75 (0.26-2.21)	−0.52	.60
History of suicide attempt	5.73 (2.07-15.90)	3.36	.001
Proband actual attempt	2.89 (1.05-7.90)	2.06	.04
Model 2
Depression symptoms class^b
2	0.45 (0.13-1.56)	−1.26	.21
3	6.74 (1.33-34.05)	2.31	.02
Class 2^c
Hopelessness	0.54 (0.14-2.09)	−0.89	.37
Impulsivity	0.72 (0.21-2.52)	−0.51	.61
Aggression	0.76 (0.25-2.31)	−0.48	.63
Impulsive aggression	2.68 (0.72-10.04)	1.47	.14
Irritability	0.37 (0.11-1.32)	−1.53	.13
Age	0.80 (0.66-0.97)	−2.24	.03
Sex, female vs male	1.48 (0.51-4.35)	0.72	.47
Race/ethnicity, white vs non-white	2.23 (0.85-5.84)	1.64	.10
Site, Pittsburgh vs New York	0.54 (0.15-1.92)	−0.95	.34
Income, class 2^c	0.74 (0.28-2.00)	−0.58	.56
Disorder
Unipolar	3.40 (0.67-17.42)	1.47	.14
Bipolar	1.01 (0.14-7.36)	0.01	.99
Anxiety	0.43 (0.11-1.78)	−1.16	.25
Alcohol and substance use	2.88 (0.79-10.49)	1.60	.11
Posttraumatic stress disorder	3.15 (0.56-17.79)	1.30	.19
Attention-deficit/hyperactivity disorder	1.35 (0.29-6.25)	0.39	.70
History
Suicide attempt	3.01 (0.86-10.49)	1.73	.08
Childhood abuse	6.99 (1.71-28.51)	2.72	.01
Proband
Actual attempt	4.70 (1.23-17.92)	2.26	.02
Bipolar disorder	0.93 (0.31-2.78)	−0.13	.90
Inpatient	3.16 (0.41-24.65)	1.10	.27
Outpatient	0.85 (0.10-7.33)	−0.15	.88
Psychotherapy	0.64 (0.12-3.56)	−0.51	.61
Antidepressant/mood stabilizer	1.92 (0.30-12.38)	0.69	.49
Other psychotropic medication	0.49 (0.09-2.77)	−0.81	.42

Open in a new tab

Abbreviation: OR, odds ratio.

^{^a}

The narrow definition included actual attempts only.

^{^b}

Dummy-coded variable with class 1 as the reference category.

^{^c}

Class 2 compared with class 1. Model 1 pseudo R² = 0.274; % correctly classified: 96.2% (full model) and 95.4% (test model). Model 2 pseudo R² = 0.392; % correctly classified: 96.8% (full model) and 94.5% (test model).

We computed a risk score using the number of statistically significant predictors endorsed from our models: younger age, in which risk was defined as 30 years or younger given that 98% of attempts occurred among offspring aged 30 years or younger; class 3 depression symptoms; having a mood disorder; history of suicide attempt; history of childhood abuse; and proband actual attempt. A risk score of 3 or higher resulted in the highest sensitivity (87.3%) and moderate specificity (63%; area under the curve = 0.80) for suicide attempt (eTable 24 in the Supplement). The positive predictive value was low but increased as the prevalence of attempt increased. Similar results were obtained for actual and incident attempts (eTable 24 in the Supplement).

Discussion

The trajectory of depression symptoms with the most severe depression symptoms and variability over time was the only trajectory to predict suicide attempt in young adults. This finding persisted even after controlling for psychiatric disorders and history of attempt. In addition, younger age, mood disorder, childhood abuse, history of suicide attempt, and proband actual attempt were statistically significant predictors. The combination of these predictors showed high sensitivity and moderate specificity.

These results highlight that the severity and variability of depression symptoms may be the only indicator of attempt above and beyond clinical characteristics. Fewer than half of offspring in class 3 depression symptoms had a diagnosis of bipolar disorder, 26 offspring (60.5%) had a diagnosis of major depressive disorder, and only 14 (6.9%) had no diagnosis of a mood disorder. Among those with a major depressive disorder diagnosis, 53.8% were younger than 25 years at the time of the last censored observation and thus were still within the age of risk for onset of bipolar disorder, had not yet fully expressed the diagnosis, or were misdiagnosed as having major depressive disorder. The mean age at onset of bipolar disorder spectrum was 14 years, with a prodrome preceding the onset ranging between 2 and 10 years.^53,54,55 These results are consistent with findings in studies showing the substantially higher risk of attempt in individuals with bipolar disorder.^56,57

A large epidemiologic community sample of adults and a meta-analysis of 27 studies with heterogeneous samples have shown depression to be a predictor associated with suicidal ideation but not attempt among those with suicidal ideation.^7,58 However, our study is based on a high-risk sample of young adult offspring of parents with mood disorders. When taking into account mood disorders and other psychiatric disorders as time-varying covariates, we found that class 3 depression symptoms continued to be associated with a 2 to 7 times increased hazard of attempt. Our results underscore the severity and variability in depression symptoms as a marker of risk for attempt.

Impulsivity and aggression have been reported to predict attempt and conceptualized as traits rather than state-dependent constructs.⁵⁹ However, we found that the trajectories of impulsivity, aggression, and impulsive aggression did not predict attempt when controlling for psychiatric disorders, and class 2 on each trajectory of impulsivity and aggression showed variability over time and thus are not stable traits. These results are also consistent with findings in a previous study showing that impulsivity is relatively state dependent and decreases in response to treatment of severe depression.⁶⁰

We identified 6 key factors (younger age, class 3 depression symptoms, mood disorder, history of suicide attempt, history of childhood abuse, and proband actual attempt) that show high sensitivity (87.3%) and moderate specificity (63%) in predicting suicide attempt. These results are consistent with screening tests that are widely used in clinical practice such as mammography, tomosynthesis, and blood-based breast cancer screening and Papanicolaou test for cervical cancer (sensitivity = 0.72-0.985; specificity = 0.4-0.67).⁶¹

The risk score we generated showed improved performance compared with other prediction models for suicide attempt and suicide risk, most of which had low sensitivity and high specificity.^{27,28,29,30,31} Weighing the cost of a false-positive against a false-negative result for an outcome like suicidal behavior in a high-risk population and the cost of screening using variables that are already collected as part of the medical history assessment in clinical settings, our risk score is a valuable addition to tools for estimating the likelihood of a suicide attempt in high-risk individuals. We acknowledge the need to test its performance in an independent sample, and future studies need to incorporate clinical, behavioral, and biological variables to achieve higher performance.⁶²

Strengths and Limitations

This study has several strengths. It is one of the few longitudinal studies to examine the predictors of suicide attempt in a well-characterized cohort with more than 12 years of follow-up. This cohort comprised a large sample of offspring at high risk for suicidal behavior. In addition, the study is one of the few studies of its kind to focus on a relatively short window of time from assessment to attempt (median time, 30-45 weeks). Simulation studies for latent class analysis have shown that a sample size of 600 or more is useful when the effect size (Cohen d) between classes is fewer than 1.^63,64 The sample size of 663 and the differences between classes were quite large (Cohen d = 2.6-7.31; Figure 1).

This study also has limitations. Participating in the study and applying our safety protocols may have reduced suicidal events among the offspring in the sample. However, we included suicide-related behaviors and conducted emergency referrals during the study (30 of 663 [4.5%] participants had referrals). With more than 150 participants per class, we were able to identify the measures on which the classes differed with large effect sizes. One case with a small class sample size was class 3 depression symptoms (n = 43). However, suicide attempt was a rare event, and as such it was not surprising that the class that predicted attempt included a small subset of the sample. We were not able to examine the trajectories of suicidal ideation because our measure was focused on intent and plan among those with suicidal ideation. As such, we did not capture the full spectrum of ideation. The study did not include a measure of mood lability to assess rapid mood variability over shorter periods. Offspring lost to follow-up were younger and showed lower rates of anxiety disorders compared with those retained for longer periods. However, we controlled for these variables in our models. The measure of irritability was limited to 1 item on depression scales. Finally, we used the same data set to estimate risk and to derive the risk score. Although machine learning methods are often used for large data sets, they are known to perform well for smaller sample sizes.^65,66 Future studies are needed to test our derived risk score in independent samples.

Conclusions

We recommend that clinicians, in their assessment of depression, (1) pay particular attention to the severity of both current and past depression and the variability in these symptoms, and (2) monitor and treat depression symptoms over time to reduce symptom severity and fluctuation, and thus the likelihood for suicide attempt, in high-risk young adults.

Supplement.

eFigure 1. Flow Diagram Describing the Sample

eFigure 2. Longitudinal Trajectories for Income Throughout the Study Period

eTable 1. Baseline Comparisons on Demographic and Clinical Characteristics of Subjects Included and Those Excluded From Analyses

eTable 2. Comparisons of Trajectories on Mean Scores of Symptoms Across Time Points

eTable 3. Concordance Between Trajectories Across Clinical Predictors

eTable 4. Demographic, Clinical, and Psychosocial Characteristics of Trajectories on Depression Symptoms

eTable 5. Demographic, Clinical, and Psychosocial Characteristics of Trajectories on Hopelessness

eTable 6. Demographic, Clinical, and Psychosocial Characteristics of Trajectories on Impulsivity

eTable 7. Demographic, Clinical, and Psychosocial Characteristics of Trajectories on Aggression

eTable 8. Demographic, Clinical, and Psychosocial Characteristics of Trajectories on Impulsive Aggression

eTable 9. Demographic, Clinical, and Psychosocial Characteristics of Trajectories on Irritability

eTable 10. Prediction of Suicide Attempt (Broad Definition) Using Individual Symptom Trajectories

eTable 11. Prediction of Actual Attempt (Narrow Definition) Using Individual Symptom Trajectories

eTable 12. Prediction of Incident Suicide Attempt (Broad Definition) Using Individual Symptom Trajectories

eTable 13. Cox Regression Models for Time to Onset of Suicide Attempt (Broad Definition) Using Individual Symptom Trajectories

eTable 14. Cox Regression Models for Time to Onset of Actual Attempt (Narrow Definition) Using Individual Symptom Trajectories

eTable 15. Cox Regression Models for Time to Onset of Incident Suicide Attempt (Broad Definition) Using Individual Symptom Trajectories

eTable 16. Sensitivity Analyses Examining the Prediction of Suicide Attempt (Broad Definition) Combining Symptom Trajectories Using the Non-imputed Dataset

eTable 17. Prediction of Suicide Attempt (Broad Definition) Combining Symptom Trajectories Among Offspring 18 Years of Age and Older

eTable 18. Sensitivity Analyses Examining the Prediction of Actual Attempt (Narrow Definition) Combining Symptom Trajectories Using the Non-imputed Dataset

eTable 19. Prediction of Incident Suicide Attempt (Broad Definition) Combining Symptom Trajectories

eTable 20. Cox Regression Models for Time to Onset of Suicide Attempt (Broad Definition) Combining Symptom Trajectories

eTable 21. Cox Regression Models for Time to Onset of Actual Attempt (Narrow Definition) Combining Symptom Trajectories

eTable 22. Cox Regression Models for Time to Onset of Incident Suicide Attempt Combining Symptom Trajectories

eTable 23. Prediction of Broad, Narrow, and Incident Suicide Attempt Definitions Combining Symptom Trajectories Using LASSO

eTable 24. Sensitivity, Specificity, Positive Predictive Value (PPV), Negative Predictive Value (NPV), and F1 Score for Various Cut-offs on the Prediction Risk Score

Click here for additional data file.^{(222.6KB, pdf)}

References

1.Murphy SL, Xu J, Kochanek KD. Deaths: final data for 2010. Natl Vital Stat Rep. 2013;61(4):1-117. [PubMed] [Google Scholar]
2.Coryell W, Young EA. Clinical predictors of suicide in primary major depressive disorder. J Clin Psychiatry. 2005;66(4):412-417. doi: 10.4088/JCP.v66n0401 [DOI] [PubMed] [Google Scholar]
3.Angst J, Angst F, Gerber-Werder R, Gamma A. Suicide in 406 mood-disorder patients with and without long-term medication: a 40 to 44 years’ follow-up. Arch Suicide Res. 2005;9(3):279-300. doi: 10.1080/13811110590929488 [DOI] [PubMed] [Google Scholar]
4.Yoshimasu K, Kiyohara C, Miyashita K; Stress Research Group of the Japanese Society for Hygiene . Suicidal risk factors and completed suicide: meta-analyses based on psychological autopsy studies. Environ Health Prev Med. 2008;13(5):243-256. doi: 10.1007/s12199-008-0037-x [DOI] [PMC free article] [PubMed] [Google Scholar]
5.Bernal M, Haro JM, Bernert S, et al. ; ESEMED/MHEDEA Investigators . Risk factors for suicidality in Europe: results from the ESEMED study. J Affect Disord. 2007;101(1-3):27-34. doi: 10.1016/j.jad.2006.09.018 [DOI] [PubMed] [Google Scholar]
6.Nock MK, Borges G, Bromet EJ, et al. Cross-national prevalence and risk factors for suicidal ideation, plans and attempts. Br J Psychiatry. 2008;192(2):98-105. doi: 10.1192/bjp.bp.107.040113 [DOI] [PMC free article] [PubMed] [Google Scholar]
7.Nock MK, Hwang I, Sampson NA, Kessler RC. Mental disorders, comorbidity and suicidal behavior: results from the National Comorbidity Survey Replication. Mol Psychiatry. 2010;15(8):868-876. doi: 10.1038/mp.2009.29 [DOI] [PMC free article] [PubMed] [Google Scholar]
8.Melhem NM, Brent DA, Ziegler M, et al. Familial pathways to early-onset suicidal behavior: familial and individual antecedents of suicidal behavior. Am J Psychiatry. 2007;164(9):1364-1370. doi: 10.1176/appi.ajp.2007.06091522 [DOI] [PMC free article] [PubMed] [Google Scholar]
9.Schaffer A, Isometsä ET, Tondo L, et al. International Society for Bipolar Disorders Task Force on Suicide: meta-analyses and meta-regression of correlates of suicide attempts and suicide deaths in bipolar disorder. Bipolar Disord. 2015;17(1):1-16.doi: 10.1111/bdi.12271 [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Brent DA, Melhem NM, Oquendo M, et al. Familial pathways to early-onset suicide attempt: a 5.6-year prospective study. JAMA Psychiatry. 2015;72(2):160-168. doi: 10.1001/jamapsychiatry.2014.2141 [DOI] [PMC free article] [PubMed] [Google Scholar]
11.Nock MK, Ursano RJ, Heeringa SG, et al. ; Army STARRS Collaborators . Mental disorders, comorbidity, and pre-enlistment suicidal behavior among new soldiers in the US Army: results from the Army Study to Assess Risk and Resilience in Servicemembers (Army STARRS). Suicide Life Threat Behav. 2015;45(5):588-599. doi: 10.1111/sltb.12153 [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Goldstein TR, Bridge JA, Brent DA. Sleep disturbance preceding completed suicide in adolescents. J Consult Clin Psychol. 2008;76(1):84-91. doi: 10.1037/0022-006X.76.1.84 [DOI] [PMC free article] [PubMed] [Google Scholar]
13.Bernert RA, Kim JS, Iwata NG, Perlis ML. Sleep disturbances as an evidence-based suicide risk factor. Curr Psychiatry Rep. 2015;17(3):554. doi: 10.1007/s11920-015-0554-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
14.Fawcett J, Busch KA, Jacobs D, Kravitz HM, Fogg L. Suicide: a four-pathway clinical-biochemical model. Ann N Y Acad Sci. 1997;836:288-301. doi: 10.1111/j.1749-6632.1997.tb52366.x [DOI] [PubMed] [Google Scholar]
15.Mann JJ. Neurobiology of suicidal behaviour. Nat Rev Neurosci. 2003;4(10):819-828. doi: 10.1038/nrn1220 [DOI] [PubMed] [Google Scholar]
16.Sareen J, Cox BJ, Afifi TO, et al. Anxiety disorders and risk for suicidal ideation and suicide attempts: a population-based longitudinal study of adults. Arch Gen Psychiatry. 2005;62(11):1249-1257. doi: 10.1001/archpsyc.62.11.1249 [DOI] [PubMed] [Google Scholar]
17.ten Have M, de Graaf R, van Dorsselaer S, et al. Incidence and course of suicidal ideation and suicide attempts in the general population. Can J Psychiatry. 2009;54(12):824-833. doi: 10.1177/070674370905401205 [DOI] [PubMed] [Google Scholar]
18.Beck AT, Steer RA, Kovacs M, Garrison B. Hopelessness and eventual suicide: a 10-year prospective study of patients hospitalized with suicidal ideation. Am J Psychiatry. 1985;142(5):559-563. doi: 10.1176/ajp.142.5.559 [DOI] [PubMed] [Google Scholar]
19.Afifi TO, Boman J, Fleisher W, Sareen J. The relationship between child abuse, parental divorce, and lifetime mental disorders and suicidality in a nationally representative adult sample. Child Abuse Negl. 2009;33(3):139-147. doi: 10.1016/j.chiabu.2008.12.009 [DOI] [PubMed] [Google Scholar]
20.Haw C, Hawton K. Life problems and deliberate self-harm: associations with gender, age, suicidal intent and psychiatric and personality disorder. J Affect Disord. 2008;109(1-2):139-148. doi: 10.1016/j.jad.2007.12.224 [DOI] [PubMed] [Google Scholar]
21.Troister T, Links PS, Cutcliffe J. Review of predictors of suicide within 1 year of discharge from a psychiatric hospital. Curr Psychiatry Rep. 2008;10(1):60-65. doi: 10.1007/s11920-008-0011-8 [DOI] [PubMed] [Google Scholar]
22.Stein DJ, Chiu WT, Hwang I, et al. Cross-national analysis of the associations between traumatic events and suicidal behavior: findings from the WHO World Mental Health Surveys. PLoS One. 2010;5(5):e10574. doi: 10.1371/journal.pone.0010574 [DOI] [PMC free article] [PubMed] [Google Scholar]
23.Bruffaerts R, Demyttenaere K, Borges G, et al. Childhood adversities as risk factors for onset and persistence of suicidal behaviour. Br J Psychiatry. 2010;197(1):20-27. doi: 10.1192/bjp.bp.109.074716 [DOI] [PMC free article] [PubMed] [Google Scholar]
24.Zatti C, Rosa V, Barros A, et al. Childhood trauma and suicide attempt: a meta-analysis of longitudinal studies from the last decade. Psychiatry Res. 2017;256:353-358. doi: 10.1016/j.psychres.2017.06.082 [DOI] [PubMed] [Google Scholar]
25.Aaltonen K, Näätänen P, Heikkinen M, et al. Differences and similarities of risk factors for suicidal ideation and attempts among patients with depressive or bipolar disorders. J Affect Disord. 2016;193:318-330. doi: 10.1016/j.jad.2015.12.033 [DOI] [PubMed] [Google Scholar]
26.Franklin JC, Ribeiro JD, Fox KR, et al. Risk factors for suicidal thoughts and behaviors: a meta-analysis of 50 years of research. Psychol Bull. 2017;143(2):187-232. doi: 10.1037/bul0000084 [DOI] [PubMed] [Google Scholar]
27.Kessler RC, Warner CH, Ivany C, et al. ; Army STARRS Collaborators . Predicting suicides after psychiatric hospitalization in US Army soldiers: the Army Study to Assess Risk and Resilience in Servicemembers (Army STARRS). JAMA Psychiatry. 2015;72(1):49-57. doi: 10.1001/jamapsychiatry.2014.1754 [DOI] [PMC free article] [PubMed] [Google Scholar]
28.Walsh CG, Ribeiro JD, Franklin JC. Predicting risk of suicide attempts over time through machine learning. Clin Psychol Sci. 2017;5(3):457-469. doi: 10.1177/2167702617691560 [DOI] [Google Scholar]
29.Barak-Corren Y, Castro VM, Javitt S, et al. Predicting suicidal behavior from longitudinal electronic health records. Am J Psychiatry. 2017;174(2):154-162. doi: 10.1176/appi.ajp.2016.16010077 [DOI] [PubMed] [Google Scholar]
30.Kessler RC, Stein MB, Petukhova MV, et al. ; Army STARRS Collaborators . Predicting suicides after outpatient mental health visits in the Army Study to Assess Risk and Resilience in Servicemembers (Army STARRS). Mol Psychiatry. 2017;22(4):544-551. doi: 10.1038/mp.2016.110 [DOI] [PMC free article] [PubMed] [Google Scholar]
31.Tran T, Luo W, Phung D, et al. Risk stratification using data from electronic medical records better predicts suicide risks than clinician assessments. BMC Psychiatry. 2014;14:76-84. doi: 10.1186/1471-244X-14-76 [DOI] [PMC free article] [PubMed] [Google Scholar]
32.Meyer RE, Salzman C, Youngstrom EA, et al. Suicidality and risk of suicide–definition, drug safety concerns, and a necessary target for drug development: a consensus statement. J Clin Psychiatry. 2010;71(8):e1-e21. doi: 10.4088/JCP.10cs06070blu [DOI] [PubMed] [Google Scholar]
33.U.S. Department of Health and Human Services, Food and Drug Administration, Center for Drug Evaluation and Research (CDER) Guidance for Industry. Suicidal ideation and behavior: prospective assessment of occurrence in clinical trials. https://www.fda.gov/drugs/guidancecomplianceregulatoryinformation/guidances/ucm315156.htm. Published August 2012. Accessed January 21, 2019.
34.Crosby A, Ortega L, Melanson C. Self-Directed Violence Surveillance: Uniform Definitions and Recommended Data Elements. Atlanta, GA: Centers for Disease Control and Prevention, National Center for Injury Prevention and Control, Division of Violence Prevention; 2011. https://www.cdc.gov/violenceprevention/pdf/Self-Directed-Violence-a.pdf. Accessed January 21, 2019. [Google Scholar]
35.Brent DA, Emslie GJ, Clarke GN, et al. Predictors of spontaneous and systematically assessed suicidal adverse events in the treatment of SSRI-resistant depression in adolescents (TORDIA) study. Am J Psychiatry. 2009;166(4):418-426. doi: 10.1176/appi.ajp.2008.08070976 [DOI] [PMC free article] [PubMed] [Google Scholar]
36.Kaufman J, Birmaher B, Brent D, et al. Schedule for Affective Disorders and Schizophrenia for School-Age Children-Present and Lifetime Version (K-SADS-PL): initial reliability and validity data. J Am Acad Child Adolesc Psychiatry. 1997;36(7):980-988. doi: 10.1097/00004583-199707000-00021 [DOI] [PubMed] [Google Scholar]
37.Spitzer RL, Williams JB, Gibbon M, First MB. The Structured Clinical Interview for DSM-III-R (SCID). I: history, rationale, and description. Arch Gen Psychiatry. 1992;49(8):624-629. doi: 10.1001/archpsyc.1992.01820080032005 [DOI] [PubMed] [Google Scholar]
38.Mann JJ, McBride PA, Brown RP, et al. Relationship between central and peripheral serotonin indexes in depressed and suicidal psychiatric inpatients. Arch Gen Psychiatry. 1992;49(6):442-446. doi: 10.1001/archpsyc.1992.01820060022003 [DOI] [PubMed] [Google Scholar]
39.Beck AT, Ward CH, Mendelson M, Mock J, Erbaugh J. An inventory for measuring depression. Arch Gen Psychiatry. 1961;4:561-571. doi: 10.1001/archpsyc.1961.01710120031004 [DOI] [PubMed] [Google Scholar]
40.Kovacs M. The Children’s Depression, Inventory (CDI). Psychopharmacol Bull. 1985;21(4):995-998. [PubMed] [Google Scholar]
41.Beck AT, Weissman A, Lester D, Trexler L. The measurement of pessimism: the Hopelessness Scale. J Consult Clin Psychol. 1974;42(6):861-865. doi: 10.1037/h0037562 [DOI] [PubMed] [Google Scholar]
42.Kazdin AE, Rodgers A, Colbus D. The Hopelessness Scale for Children: psychometric characteristics and concurrent validity. J Consult Clin Psychol. 1986;54(2):241-245. doi: 10.1037/0022-006X.54.2.241 [DOI] [PubMed] [Google Scholar]
43.Barratt ES. Factor analysis of some psychometric measures of impulsiveness and anxiety. Psychol Rep. 1965;16:547-554. doi: 10.2466/pr0.1965.16.2.547 [DOI] [PubMed] [Google Scholar]
44.Plomin R. Childhood temperament In: Lahey BB, Kazdin AE, eds. Advances in Clinical Child Psychology. New York: Plenum; 1983:45-92. doi: 10.1007/978-1-4613-9814-1_2 [DOI] [Google Scholar]
45.Buss AH, Durkee A. An inventory for assessing different kinds of hostility. J Consult Psychol. 1957;21(4):343-349. doi: 10.1037/h0046900 [DOI] [PubMed] [Google Scholar]
46.Kazdin AE, Rodgers A, Colbus D, Siegel T. Children’s Hostility Inventory: measurement of aggression and hostility in psychiatric inpatient children. J Clin Child Psychol. 1987;16:320-328. doi: 10.1207/s15374424jccp1604_5 [DOI] [Google Scholar]
47.Brown GL, Goodwin FK. Human aggression and suicide. Suicide Life Threat Behav. 1986;16(2):223-243. doi: 10.1111/j.1943-278X.1986.tb00353.x [DOI] [PubMed] [Google Scholar]
48.Bernstein DP, Fink L, Handelsman L, et al. Initial reliability and validity of a new retrospective measure of child abuse and neglect. Am J Psychiatry. 1994;151(8):1132-1136. doi: 10.1176/ajp.151.8.1132 [DOI] [PubMed] [Google Scholar]
49.Chaffin M, Wherry JN, Newlin C, Crutchfield A, Dykman R. The Abuse Dimensions Inventory: initial data on a research measure of abuse severity. J Interpers Violence. 1997;12(4):569-589. doi: 10.1177/088626097012004006 [DOI] [Google Scholar]
50.Stata Statistical Software [computer program]. Release 11. College Station, TX: StataCorp LLC; 2009.
51.Collins LM, Schafer JL, Kam CM. A comparison of inclusive and restrictive strategies in modern missing data procedures. Psychol Methods. 2001;6(4):330-351. doi: 10.1037/1082-989X.6.4.330 [DOI] [PubMed] [Google Scholar]
52.Friedman J, Hastie T, Tibshirani R. Regularization paths for generalized linear models via coordinate descent. J Stat Softw. 2010;33(1):1-22. doi: 10.18637/jss.v033.i01 [DOI] [PMC free article] [PubMed] [Google Scholar]
53.Hafeman DM, Merranko J, Goldstein TR, et al. Assessment of a person-level risk calculator to predict new-onset bipolar spectrum disorder in youth at familial risk. JAMA Psychiatry. 2017;74(8):841-847. doi: 10.1001/jamapsychiatry.2017.1763 [DOI] [PMC free article] [PubMed] [Google Scholar]
54.Van Meter AR, Burke C, Youngstrom EA, Faedda GL, Correll CU. The bipolar prodrome: meta-analysis of symptom prevalence prior to initial or recurrent mood episodes. J Am Acad Child Adolesc Psychiatry. 2016;55(7):543-555. doi: 10.1016/j.jaac.2016.04.017 [DOI] [PubMed] [Google Scholar]
55.Conus P, Ward J, Hallam KT, et al. The proximal prodrome to first episode mania–a new target for early intervention. Bipolar Disord. 2008;10(5):555-565. doi: 10.1111/j.1399-5618.2008.00610.x [DOI] [PubMed] [Google Scholar]
56.Pompili M, Gonda X, Serafini G, et al. Epidemiology of suicide in bipolar disorders: a systematic review of the literature. Bipolar Disord. 2013;15(5):457-490. doi: 10.1111/bdi.12087 [DOI] [PubMed] [Google Scholar]
57.Grande I, Berk M, Birmaher B, Vieta E. Bipolar disorder. Lancet. 2016;387(10027):1561-1572. doi: 10.1016/S0140-6736(15)00241-X [DOI] [PubMed] [Google Scholar]
58.May A, Klonsky ED. What distinguishes suicide attempters from suicide ideators? a meta-analysis of potential factors. Clin Psychol Sci Pract. 2016;23(1):5-20. doi: 10.1111/cpsp.12136 [DOI] [Google Scholar]
59.Mann JJ, Arango VA, Avenevoli S, et al. Candidate endophenotypes for genetic studies of suicidal behavior. Biol Psychiatry. 2009;65(7):556-563. doi: 10.1016/j.biopsych.2008.11.021 [DOI] [PMC free article] [PubMed] [Google Scholar]
60.Corruble E, Benyamina A, Bayle F, Falissard B, Hardy P. Understanding impulsivity in severe depression? a psychometrical contribution. Prog Neuropsychopharmacol Biol Psychiatry. 2003;27(5):829-833. doi: 10.1016/S0278-5846(03)00115-5 [DOI] [PubMed] [Google Scholar]
61.Maxim LD, Niebo R, Utell MJ. Screening tests: a review with examples. Inhal Toxicol. 2014;26(13):811-828. doi: 10.3109/08958378.2014.955932 [DOI] [PMC free article] [PubMed] [Google Scholar]
62.Melhem NM, Munroe S, Marsland A, et al. Blunted HPA axis activity prior to suicide attempt and increased inflammation in attempters. Psychoneuroendocrinology. 2017;77:284-294. doi: 10.1016/j.psyneuen.2017.01.001 [DOI] [PMC free article] [PubMed] [Google Scholar]
63.Dziak JJ, Lanza ST, Tan X. Effect size, statistical power and sample size requirements for the bootstrap likelihood ratio test in latent class analysis. Struct Equ Modeling. 2014;21(4):534-552. doi: 10.1080/10705511.2014.919819 [DOI] [PMC free article] [PubMed] [Google Scholar]
64.Gudicha DW, Tekle FB, Vermunt JK. Power and sample size computation for Wald tests in latent class models. J Classif. 2016;33(1):30-51. doi: 10.1007/s00357-016-9199-1 [DOI] [Google Scholar]
65.Hastie T, Tibshirani R, Friedman J. The Elements of Statistical Learning: Data Mining, Inference, and Prediction. 2nd ed New York: Springer; 2009. doi: 10.1007/978-0-387-84858-7 [DOI] [Google Scholar]
66.Izenman AJ. Modern Multivariate Statistical Techniques: Regression, Classification, and Manifold Learning. New York: Springer; 2013. [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials