Evaluating the Cross-Sectional and Longitudinal Relationships Predicting Suicidal Ideation Following Traumatic Brain Injury

Abstract

Objective:

Characterize relationships among substance misuse, depression, employment, and suicidal ideation (SI) following moderate to severe traumatic brain injury (TBI).

Design:

Prospective cohort study.

Setting:

Inpatient rehabilitation centers with telephone follow-up; level I/II trauma centers in the United States.

Participants:

Individuals with moderate to severe TBI with data in both the National Trauma Data Bank and the Traumatic Brain Injury Model Systems National Database, aged 18 to 59 years, with SI data at year 1 or year 2 postinjury (N = 1377).

Main Outcome Measure:

Primary outcome of SI, with secondary employment, substance misuse, and depression outcomes at years 1 and 2 postinjury.

Results:

Cross-lagged structural equation modeling analysis showed that year 1 unemployment and substance misuse were associated with a higher prevalence of year 1 depression. Depression was associated with concurrent SI at years 1 and 2. Older adults and women had a greater likelihood of year 1 depression. More severe overall injury (injury severity score) was associated with a greater likelihood of year 1 SI, and year 1 SI was associated with a greater likelihood of year 2 SI.

Conclusions:

Substance misuse, unemployment, depression, and greater extracranial injury burden independently contributed to year 1 SI; in turn, year 1 SI and year 2 depression contributed to year 2 SI. Older age and female sex were associated with year 1 depression. Understanding and mitigating these risk factors are crucial for effectively managing post-TBI SI to prevent postinjury suicide.

TRAUMATIC BRAIN INJURY (TBI) is a complex injury that can lead to the development of multiple chronic conditions, including mood disorders, and associated disability.^1,2 Traumatic brain injury is associated with high rates of unemployment^3–5 and substance misuse,^6–8 each of which may have a bidirectional relationship with mood disorders^5,9,10 and suicidal ideation (SI).¹¹

Some studies report a 2 -to-4-fold increased risk of death by suicide among individuals with TBI compared with the general population.^11–14 A study of 172 individuals with TBI reported that 23% experienced post-TBI SI and 17% attempted suicide,¹⁵ nearly 3 times the general trauma population rate.¹⁶ Chronic TBI sequelae, including chronic pain,¹⁷ disinhibition,¹⁸ and negative affect,¹⁸ are known predictors of suicide attempts and SI in the general population. Post-TBI suicide risk often co-occurs with substance misuse,¹⁴ poor cognitive recovery,¹⁹ anxiety,^19,20 and depression.^13,21

Post-TBI employment is an indicator of rehabilitation and disability mitigation. Employment can enhance quality of life through improved community reintegration and financial independence.^3,4,22,23 However, individuals post-TBI experience increased unemployment and employment instability up to 10 years post-TBI.^24–27 Also, associations exist between substance misuse (including drug and alcohol abuse) and psychological, behavioral, and functional outcomes post-TBI, including employment.^9,28,29 Our recent study demonstrated a relationship between substance misuse and later employment post-TBI.³⁰ Traumatic brain injury is associated with 8 times higher likelihood of depression during the first year postinjury compared with the general population.³¹ Depression is also associated with cognitive impairment,³² reduced function,³³ postinjury unemployment,³⁴ and SI.²¹

Our previous work evaluated the relationship between extracranial injury severity score (ISS) and SI.³⁵ This study identified severe extracranial injury (ECI) as associated with SI across the first 5 years post-TBI while controlling for relevant covariates, including substance misuse and depression. Additional studies involving general trauma populations and spinal cord injury suggest elevated risk of SI,³⁶ with predictors including chronic pain,^16,17 low community participation,³⁷ substance misuse,³⁸ and depression.³⁷ As a result, we hypothesized a conceptual pathway through which the ECI complex impacts SI through substance misuse, depression, and employment.

In our previous work, we used cross-lagged structural equation modeling (CLSEM) to identify the longitudinal interrelationships between employment and substance misuse. Cross-lagged structural equation modeling allows for the simultaneous evaluation of outcomes across follow-up time points. We found that ECI was associated with lower likelihood of employment and higher likelihood of substance misuse, as hypothesized. However, employment was also associated with subsequent substance misuse.³⁰ We supposed that employed individuals post-TBI could be using substances to cope with associated stress of their employment, and that future work needed to investigate concurrent emotional and mental health factors. While employment, substance misuse, depression, and SI share common predictors and have exhibited relationships with one another, no previous study (1) has characterized the relationships among these variables or (2) used these relationships to characterize SI risk post-TBI.

Our objective was to characterize factors affecting post-TBI SI, including static predictors (age, sex, and ECI burden measured by ISS) and longitudinal interrelationships between SI and employment, substance misuse, and depression at 1 and 2 years post-TBI.

METHODS

Participants

This analysis studied individuals with moderate to severe TBI, enrolled in the TBI Model Systems (TBIMS) study, with data collected from emergency, acute, and rehabilitation care as well as via follow-up telephone interviews at 1, 2, and 5 years and every subsequent 5 years. Traumatic Brain Injury Model Systems centers locally enrolled participants with moderate to severe TBI who were 16 years of age or older at the time of injury, presenting to a TBIMS acute care hospital within 72 hours of injury, and receiving inpatient rehabilitation at a TBIMS facility.³⁹ All centers follow site-specific institutional review board approved protocols and obtain informed consent.

Traumatic Brain Injury Model Systems participants were previously linked with their respective trauma records from the National Trauma Data Bank (NTDB) and local trauma registries.^40,41 The NTDB is the largest aggregation of trauma data in the United States (https://www.facs.org/quality-programs/trauma/tqp/center-programs/ntdb). These trauma databases provide information beyond that captured in the TBIMS concerning acute injury complex and hospital course. In 2 centers with available personal identifiers, the TBIMS participants were deterministically linked with their local trauma center’s registry records. Data from study participants enrolled at other TBIMS centers, with no available identifiers, were probabilistically linked with their NTDB records. This linking process had been previously described, validated, and analyzed.^35,40–42 Injury severity score was extracted from the NTDB and all other data from the TBIMS national database.

We restricted this study to participants aged 18 to 59 years at injury, given our study’s focus on employment, mirroring previous studies evaluating employment^27,30 and as employment is less likely beyond this age range.³ We also required available SI data at year 1 or year 2. Finally, inclusion in structural equation model (SEM) or logistic regression modeling was additionally limited to individuals with all nonmissing outcome data.^43,44 Sample selection is summarized in Figure 1. We compared characteristics of individuals included in CLSEM analysis with those excluded because of missing data.

Figure 1:

Inclusion and exclusion criteria for study cohort, cross-sectional SEM, and CLSEM

Flow chart of exclusion criteria and included individuals. Within cohort of individuals matched across the TBIMS and NTDB databases, 2,813 individuals with ISS and outcome data who were between 18- and 59- years old at time of injury and who were not students at any point within 2-years of injury were eligible for inclusion in this study. Of those 2,813 individuals, 1,039 and 958 were included in cross-sectional SEM of SE at years-1 and -2 respectively. Finally, 716 individuals were included in a CLSEM model of employment, depression, SA, and SE, which included year-1 and -2 outcomes.

Abb: SEM=Structural Equations Model; CLSEM=Cross-Lagged SEM; SA = Substance Abuse; SE = Suicidal Endorsement; ISS = Injury Severity Score

Outcomes

Outcome measures were collected at years 1 and 2 postinjury. Self-reported SI was the primary outcome, with self-reported employment, substance misuse, and depression as secondary outcomes. Employment status is a binary variable reflecting competitive or special employment (full time/part time) versus unemployment (including unemployment, early retirement, volunteer work, leave, or primarily taking care of the household).

Suicidal ideation was categorized as “present” or “absent” based on item 9 of the PHQ-9, asking: “Over the last two weeks, how often have you been bothered by thoughts that you would be better off dead or of hurting yourself in some way?” Suicidal ideation was defined as present for participants reporting the symptom present at all over the last 2 weeks (item score ≥1). Item 9 of the PHQ-9 questionnaire has previously been used to identify SI in individuals with TBI.^45,46

Substance misuse was categorized as “present” or “absent.” Present substance misuse during a follow-up interview was indicated by (1) use of any illicit drug; (2) any binge drinking during the last month (defined as ≥5 drinks at once); or (3) more than 14 drinks per week in men, 7 in women).^47,48

Depression was defined as “present” or “absent” using the following validated criterion: at least 5 symptoms present from the first 8 questions of the PHQ-9 for at least several days over the last 2 weeks, with at least one of the symptoms being anhedonia or depressed mood.⁴⁹ As PHQ item 9 determined SI presence, we omitted this item when determining depression status.

Covariates

We included age, sex, and ISS as primary predictors of post-TBI SI based on previous studies.^14,35,50 Injury severity score is a clinically assessed, anatomic trauma severity scale that calculates the sum of the squared Abbreviated Injury Scale scores of the 3 most severely injured body regions to quantify trauma severity.⁵¹ Injury severity score values range from 1 to 75, with higher scores indicating increased severity.

Regression and SEM models were controlled for preinjury marital status (single; married; divorced/ widowed), education (<12 years; high school diploma; some college or more), preinjury employment, preinjury substance misuse, rehospitalization within year prior to follow-up (yes/no), TBI severity, and social support. Traumatic brain injury severity was categorized as severe for Glasgow Coma Scale⁵² scores of less than 9, posttraumatic amnesia more than 1 week, or inability to follow verbal commands for more than 1 day, and otherwise as moderate.⁵³ The “social relations” domain of the Participation Assessment with Recombined Tools (PART) measure, ranging from 0 (lowest level of social participation) to 5 (highest level) measured “social suppot.”^54,55

Statistical analysis

Data were summarized as mean ± SD, median (interquartile range), or frequency (percentage) as applicable. We characterized the population and tested significant differences across SI prevalence via χ² test for categorical variables and Wilcoxon rank sum or 2-sample t test as applicable for continuous variables, with a significance level of .05.

Eight logistic regression models, evaluating associated predictors of employment, substance misuse, depression, and SI at year 1 and year 2, respectively, informed SEM analysis. Models included primary predictors of age, sex, ISS, concurrent employment, substance misuse, and depression variables (excluding the outcome variable itself). Models with year 2 outcomes as the dependent variable additionally included year 1 status of the same variable as a covariate. For example, a model with year 2 depression as the outcome included predictors of age, sex, ISS, year 2 employment and substance misuse, and year 1 depression.

Cross-sectional SEM was then conducted at year 1 and year 2 postinjury to test hypothesized pathways through age, sex, ISS, employment, substance misuse, and depression to SI. We used a confounder adjusted CLSEM modeling approach to support assessment of temporal relationships among employment, substance misuse, depression, and SI, using the diagonally weighted least squares estimation method. The CLSEM analyzed all year 1 and year 2 variables simultaneously to produce an adjusted conditional model for lagged effects, efficiently analyzing multiple, temporally dependent relationships. The relationships measured in CLSEM are graphically summarized in Figure 2, with unidirectional arrows representing probit regressions. Based on recommended guidelines, we assessed model fit for each CLSEM via the root-mean-square error of approximation, comparative fit index, and Tucker-Lewis index.^56,57 As the cross-sectional SEMs were just identified,⁵⁸ satisfactory standardized root-mean-square residuals assessed model fit in place of root-mean-square error of approximation, comparative fit index, and Tucker-Lewis index. Unstandardized path coefficients (β) are presented in SEM and CLSEM results, with standardized path coefficients and full SEM/CLSEM results available in Supplemental Digital Content Tables 3 to 5, available at: http://links.lww.com/JHTR/A344. Standardized path coefficients (β) greater than 0.50 indicate a large effect, near 0.30 a medium effect, and 0.10 a small effect.⁵⁹

Figure 2:

Hypothesized SEM for year 1 and 2 cross-sectional and CLSEM analyses

Conceptual figure showing the relationships tested by cross-sectional SEM (denoted by solid arrows) and relationships tested in CLSEM analysis (denoted by solid arrows and dashed arrows).

Abb: SEM = Structural Equation Model; CLSEM = Cross-Lagged SEM

Post hoc analyses included using χ² or Fisher exact tests to compare SI prevalence across preinjury mental health treatment, suicide attempt, and psychiatric hospitalization (all in year prior to injury) and across violent compared with nonviolent injury etiology. Violent injuries included gunshot wounds, assaults, or other violent injury mechanisms (eg, blasts). These variables were not included in modeling due to high levels of incomplete data. All analyses were conducted in R, with SEM conducted using the lavaan package.^60,61

RESULTS

Participant characteristics

Sample characteristics for the study cohort and by SI status are presented in Table 1. Compared with individuals without SI, individuals with SI at year 1 were more often women, divorced/widowed, unemployed preinjury and at year 1, rehospitalized within their first year postinjury, had a preinjury history of substance misuse, experienced concurrent depression, and had less social support. We observed similar differences by SI status at year 2, excluding female sex and higher ISS. We observed higher rates of year 1 substance misuse, unemployment, and SI among individuals with year 2 SI. Of individuals with depression at year 1, more than 31% and of those with depression at year 2, more than 29% ideated suicidal thoughts. More than 10% of individuals who ideated suicidal thoughts did not concurrently meet criteria for depression at both year 1 (n = 16, 11.43%) and year 2 (n = 15, 12.43%). Individuals included in the CLSEM did not differ from those excluded for missing data (see Supplemental Digital Content Table 1, available at: http://links.lww.com/JHTR/A344).

Table 1:

Descriptive statistics by suicidal endorsement

Year 1
Covariate	Total (N=1,136)	Didn’t endorse suicide (n=994)	Endorsed suicide (n=142)	P-value
Age at injury, median [Q1-Q3]	35 [24–47]	35 [24–47]	39.5 [27–49]	0.056†
Sex (Male), N (%)	1947 (76.47)	769 (77.4)	97 (68.3)	0.023*
Injury severity scale (ISS), mean ± SD	25.67 ± 11.55	26.33 ± 11.38	28.31 ± 13.26	0.093†
Employment status at year 1 (Employed), N (%)	834 (32.96)	393 (39.7)	26 (18.3)	<0.001*
Substance abuse at year 1 (Yes), N (%)	526 (20.66)	207 (21.0)	39 (28.5)	0.061*
Depression at year 1 (Yes), N (%)	374 (33.19)	250 (25.3)	116 (82.9)	<0.001*
Marital status (Married), N (%)	622 (33.77)	331 (33.3)	44 (31.0)	0.047*
Marital status (Divorced/Widowed), N (%)	347 (18.84)	186 (18.7)	39 (27.5)	0.047*
Years of education (<=11 years), N (%)	762 (32.00)	288 (29.1)	47 (33.1)	0.183*
Years of education (‘HS diploma’), N (%)	719 (30.20)	309 (31.3)	40 (28.2)	0.588*
Pre-injury substance abuse (Yes), N (%)	568 (22.60)	192 (19.5)	47 (33.3)	<0.001*
Rehospitalization (Yes), N (%)	681 (26.75)	200 (20.1)	40 (28.2)	0.037*
TBI severity (Severe), N (%)	2,157 (85.97)	831 (83.6)	118 (83.7)	1.000*
Pre-injury employment (Yes), N (%)	1724 (75.81)	746 (78.5)	92 (66.7)	0.003*
PART Social subscale, mean ± SD	2.26 ± 0.98	2.39 ± 0.91	2.10 ± 0.95	0.001†
Year 2
Covariate	Total (N=1,034)	Didn’t endorse suicide (n=914)	Endorsed suicide (n=120)	P-value
Age at injury, median [Q1-Q3]	35 [24–47]	35 [24–47]	38.5 [26–47.25]	0.093†
Sex (Male), N (%)	1,782 (76.25)	689 (75.4)	90 (75.0)	1.000*
Injury severity scale (ISS), mean ± SD	25.67 ± 11.55	27.10 ± 11.64	28.42 ± 13.28	0.300†
Employment status at year 1 (Employed), N (%)	834 (32.96)	353 (40.5)	16 (13.8)	<0.001*
Employment status at year 2 (Employed), N (%)	868 (36.64)	397 (43.6)	17 (14.3)	<0.001*
Substance abuse at year 1 (Yes), N (%)	526 (20.66)	177 (20.6)	22 (19.3)	0.836*
Substance abuse at year 2 (Yes), N (%)	519 (22.21)	218 (24.0)	41 (34.2)	0.022*
Depression at year 1 (Yes), N (%)	374 (33.19)	185 (26.4)	61 (70.9)	<0.001*
Depression at year 2 (Yes), N (%)	337 (32.72)	233 (25.6)	96 (80.7)	<0.001*
Suicidal endorsement year 1 (Yes), N (%)	142 (12.50)	54 (7.66)	39 (44.33)	<0.001*
Marital status (Married), N (%)	793 (33.95)	304 (33.3)	41 (34.2)	0.074*
Marital status (Divorced/Widowed), N (%)	418 (17.89)	164 (18.0)	31 (25.8)	0.074*
Years of education (<=11 years), N (%)	673 (30.83)	246 (27.1)	42 (35.0)	0.122*
Years of education (‘HS diploma’), N (%)	683 (31.29)	283 (31.2)	38 (31.7)	0.122*
Pre-injury substance abuse (Yes), N (%)	507 (22.03)	186 (20.6)	33 (27.5)	0.107*
Rehospitalization (Yes), N (%)	594 (26.93)	156 (17.1)	42 (35.0)	<0.001*
TBI severity (Severe), N (%)	1,986 (86.16)	763 (83.6)	105 (88.2)	0.240*
Pre-injury employment (Yes), N (%)	1,597 (76.67)	697 (79.1)	83 (70.9)	0.059*
PART Social subscale, mean ± SD	2.25 ± 0.99	2.35 ± 0.94	2.04 ± 0.94	0.001†

Column percentages reported

^*For categorical variables: p-value was calculated using Chi-square test

^†For continuous variables ‘Age at Injury’ and ‘PART Social subscale’ p-value was calculated using two-sample Wilcoxon rank-sum (Mann-Whitney) test and for ‘Injury Severity Score’ p-value was calculated using two-sample T-test

Modeling results

Logistic regression modeling results are included in Supplemental Digital Content Table 2, available at: http://links.lww.com/JHTR/A344. Year 1 and year 2 cross-sectional, unstandardized SEM coefficients are summarized in Figures 3a and b, respectively. Full SEM results, including standardized coefficients, are available for year 1 and year 2 models in Supplemental Digital Content Tables 3 and 4, available at: http://links.lww.com/JHTR/A344, respectively. The SEM findings largely confirmed significant relationships identified with regression models.

Figure 3a:

Path diagram for year 1 SEM

Figures 3a and 3b display results of cross-sectional, SEM for depression, employment, SA, and SE at year-1 (n=1,039) and year-2 (n=958) post-injury respectively. Arrows represent significant associations (p≤0.05). Numbers represent β-coefficients from probit regressions with SEM. All models were adjusted for marital status, education, pre-injury employment, pre-injury SA, rehospitalization, social support (measured via PART-O), and TBI severity.

Abb: SEM = Structural Equation Model; SA = Substance Abuse; SE = Suicidal Endorsement; PART-O = Participation Assessment with Recombined Tools-Objective; TBI = Traumatic Brain Injury

Here, older age was associated with lower likelihood of employment (β = −.011, P = .019) and substance misuse (β = −.018, P < .001) in the year 1 cross-sectional SEM. Employment was associated with lower likelihood of depression (β = −.372, P < .001), while substance misuse was associated with higher likelihood (β = .186, P = .002) of depression. Men also had lower likelihood of depression (β = −.306, P = .002). Depression (β = .565, P < .001) and higher ISS (β = .009, P = .035) were associated with higher likelihood of SI in the year 1 cross-sectional SEM.

Year 2 cross-sectional SEM results mirrored most year 1 relationships. Exceptions included employment associations with lower likelihood of SI (β = −.203, P = .017) but greater likelihood of substance misuse (β = .124, P = .046). Depression (β = .448, P < .001) and substance misuse (β = .168, P = .015) were associated with higher likelihood of SI. Older age was associated with lower likelihood of employment (β = −.015, P = .003) and substance misuse (β = −.014, P = .012). As in year 1, men had higher likelihood of employment (β = .328, P = .004) and substance misuse (β = .238, P = .049) and lower likelihood of depression (β = −.211, P = .035). Similar to year 1, depression was also associated with employment (β = −.426, P < .001) and substance misuse (β = .202, P = .001). Injury severity score was not independently associated with year 2 SI.

The CLSEM results and unstandardized coefficients are summarized in Figure 4. Year 1 depression was asso ciated positively with older age, female sex, and year 1 substance misuse (β_YR1 = .212, P = .002) and negatively with year 1 employment (β_YR1 = −.379, P < .001). Year 2 depression was positively associated with year 1 depression (β = .587, P < .001) and had no associations with year 2 employment or substance misuse. Year 2 employment and substance misuse also had positive and significant associations with their year 1 statuses (β = .896, P < .001 and β = .661, P < .001, respectively). A higher likelihood of depression was associated with a higher likelihood of concurrent SI at year 1 (β = .567, P < .001) and year 2 (β = .263, P = .002). Suicidal ideation at year 1 was associated with higher ISS (β = .015, P = .015), and year 2 SI was associated with year 1 SI (β = .237, P = .050). No cross-lag relationships (year 1 to year 2 relationships across different outcomes, eg, year 1 depression to year 2 SI) were statistically significant. Full CLSEM results, including standardized coefficients, are presented in Supplemental Digital Content Table 5, available at: http://links.lww.com/JHTR/A344.

Figure 4:

Path diagram for year-1 and -2 CLSEM

Figures 4 displays results of CLSEM for depression, employment, SA, and SE at year-1 and year-2 (n=716) post-injury respectively. Arrows represent significant associations (p≤0.05). Numbers represent β-coefficients from probit regressions with SEM. All models were adjusted for marital status, education, pre-injury employment, pre-injury SA, rehospitalization, social support (measured via PART-O), and TBI severity.

Abb: CLSEM = Cross-Lagged Structural Equation Model; SA = Substance Abuse; SE = Suicidal Endorsement; PART-O = Participation Assessment with Recombined Tools-Objective; TBI = Traumatic Brain Injury

Post hoc analyses

Prevalence of year 1 SI was significantly higher among individuals with prior psychiatric hospitalization, previous suicide attempt, or prior mental health treatment preinjury and specifically in the year prior to TBI (see Table 2). We observed no differences in year 1 SI prevalence when compared with suicide attempt in the year prior to TBI, violent versus nonviolent injury etiology, or mental health treatment in the year prior to TBI.

Table 2:

Logistic regression analyses

Covariates	Employment			Substance abuse
	OR	P(>|z|)	[95% CI]	OR	P(>|z|)	[95% CI]
Year 1	N=1080			N=1080
Age	0.986	0.084	[0.970, 1.002]	0.967	<0.001	[0.948, 0.983]
Sex (Male)	1.073	0.706	[0.746, 1.545]	1.567	0.033	[1.047, 2.389]
Injury severity scale (ISS)	0.990	0.151	[0.977, 1.003]	0.990	0.174	[0.976, 1.004]
Employment at year 1	-	-	-	1.301	0.052	[0.998, 1.699]
Substance abuse at year 1	1.330	0.042	[1.011, 1.751]	-	-	-
Depression at year 1	0.458	<0.001	[0.356, 0.585]	1.445	0.004	[1.127, 1.850]
Year 2	N=914			N=901
Age	0.983	0.143	[0.961, 1.006]	0.979	0.040	[0.960, 0.999]
Sex (Male)	1.782	0.027	[1.073, 3.001]	1.438	0.107	[0.932, 2.259]
Injury severity scale (ISS)	0.999	0.959	[0.982, 1.019]	0.993	0.379	[0.977, 1.008]
Employment at year 1	11.450	<0.001	[8.339, 16.007]	-	-	-
Employment at year 2	-	-	-	1.439	0.018	[1.066, 1.950]
Substance abuse at year 1	-	-	-	4.191	<0.001	[3.193, 5.534]
Substance abuse at year 2	1.221	0.281	[0.847, 1.758]	-	-	-
Depression at year 1	-	-	-	-	-	-
Depression at year 2	0.446	<0.001	[0.313, 0.628]	1.581	0.001	[1.192, 2.100]
	Depression			Suicidal endorsement
	OR	P(>|z|)	[95% CI]	OR	P(>|z|)	[95% CI]
Year 1	N=1080			N=1039
Age	1.009	0.210	[0.995, 1.024]	1.013	0.243	[0.991, 1.036]
Sex (Male)	0.593	0.002	[0.428, 0.822]	0.576	0.023	[0.359, 0.930]
Injury severity scale (ISS)	1.004	0.555	[0.992, 1.016]	1.022	0.015	[1.004, 1.041]
Employment at year 1	0.460	<0.001	[0.360, 0.586]	0.717	0.106	[0.475, 1.066]
Substance abuse at year 1	1.423	0.005	[1.111, 1.822]	1.307	0.139	[0.912, 1.857]
Depression at year 1	-	-	-	5.824	<0.001	[4.141, 8.423]
Year 2	N=727			N=734
Age	1.014	0.176	[0.994, 1.035]	1.001	0.965	[0.972, 1.029]
Sex (Male)	0.737	0.194	[0.466, 1.171]	0.981	0.955	[0.522, 1.895]
Injury severity scale (ISS)	0.997	0.717	[0.980, 1.014]	1.011	0.342	[0.988, 1.036]
Employment at year 1	-	-	-	-	-	-
Employment at year 2	0.474	<0.001	[0.336, 0.662]	0.705	0.192	[0.411, 1.181]
Substance abuse at year 1	-	-	-	-	-	-
Substance abuse at year 2	1.428	0.036	[1.024, 1.992]	1.626	0.027	[1.053, 2.499]
Depression at year 1	4.778	<0.001	[3.626, 6.341]
Depression at year 2	-	-	-	3.939	<0.001	[2.557, 6.278]
Suicidal endorsement at year 1	-	-	-	3.019	<0.001	[1.954, 4.683]

Abbreviated model presented. Logistic regression was controlled for marital status, education, pre-injury drug use, post-injury rehospitalization, injury severity, social support, and pre-injury employment

TBI=Traumatic Brain Injury; CI=Confidence Interval; OR=Odds Ratio

DISCUSSION

In view of the high prevalence of post-TBI suicide compared with the general population,^12,13 there is a clear need to identify and mitigate factors leading to posttraumatic SI, with an end goal of preventing suicide post-TBI. Building upon the body of current research identifying risk factors of SI,^{11,14,16–21,62,63} and previous studies identifying relationships between ECI and SI and between ECI, employment, and substance misuse,^30,35 we identified demographic and clinical factors influencing mental health outcome and SI post-TBI. The present study uses cross-sectional and longitudinal modeling to identify factors related to SI following moderate to severe TBI. Our results unsurprisingly identify substance misuse, unemployment, and depression, and notably implicate greater ECI burden, as contributing factors to SI at year 1 post-TBI.

Our work also indicates early employment and substance misuse as factors affecting mental health and subsequent SI post-TBI. These results support the importance of rehabilitation to optimize vocational goals and screen for substance misuse, as unemployment and substance misuse are important risk factors for depression in the first year after injury. While depression may often be more proximal to SI, early vocational support or substance misuse screening may also be important preventative measures, in addition to depression screening to improve long-term mental health and reduce SI postinjury.

The implications of this independent association between ECI and SI are of tremendous clinical and public health significance and have been similarly observed in general trauma populations.^38,64 Not surprisingly, depression was a strong, independent predictor of SI, confirming the importance of detection and need for effective depression prevention and treatment after TBI. One study documented that fewer than half of individuals with self-reported depression in the first year following TBI received depression treatment.³¹ This treatment gap may reflect low depression screening rates and a lack of evidence-based depression treatment guidelines for persons with TBI.^65–67 Nonetheless, our findings suggest that (1) identifying and monitoring individuals with SI risk factors (eg, history of SI or mental health treatment, more severe ECI) and (2) early detection and mitigation of modifiable risk factors along the pathway toward SI (eg, depression, substance misuse) may decrease SI prevalence following TBI. Suicidal ideation screening should be routine with depression screening, particularly as our data show that approximately 1 in 3 depressed individuals also ideated suicide. Recent studies and expert consensus suggest that a proactive approach to identifying high-risk individuals⁶⁸ and a person-centered, flexible approach to depression treatment (including use of smart phone and other point-of-care technologies) may optimize clinical efficiency and reach.⁶⁹ As ISS predicted SI at year 1 post-TBI independently of depression, clinicians must be aware that individuals who experience severe ECI may ideate suicidal thoughts, even without endorsing other (commonly concurrent) depressive symptoms. Our data demonstrated that nearly 1 in 9 individuals endorsing suicidal thoughts did not meet diagnostic criteria for depression, reinforcing the importance of screening for SI in all individuals with moderate to severe TBI, not just those with a likely depression disorder.⁷⁰

The current study builds upon our prior work documenting a strong association between severity of ECI and SI³⁵ by adjusting for potential confounders of this association, including employment, substance misuse, social participation, and depressive symptoms. Contrary to our expectations, ECI’s effects on SI were not explained by its effects on employment, substance misuse, participation, or depression. A recent epidemiological study using the TBIMS documented that approximately 1 in 10 individuals with TBI will report SI up to 20 years postinjury, with 1 in 100 attempting suicide.⁷¹

There are several potential explanations for the heightened suicidality risk after TBI among those with ECI. Executive dysfunction frequently occurs as a result of TBI, often characterized by impulsive behaviors,⁷² which are linked to suicide attempt and ideation in uninjured populations.^73,74 Individuals who sustain ECI may represent a subpopulation of TBI more prone to risk-taking behaviors.⁷⁵ Impulsivity and resulting risk-taking behaviors may represent an important preinjury risk factor that predisposes individuals to both polytrauma and resulting postinjury SI.^76–78 Individuals with higher preinjury impulsivity may experience ECI as well as increased disinhibition and impulsivity postinjury, leading to elevated SI rates in a TBI population sustaining concurrent severe ECI, compared with less severe or no ECI.

Furthermore, individuals with polytrauma resulting from risk-taking behaviors (eg, drinking and driving, unhelmeted motorcycle crash) should be carefully screened for psychiatric comorbidities to address increased risk of reinjury. Ecological momentary assessment and related telehealth may present feasible options to carefully and consistently screen emotional and behavioral symptoms chronically post-TBI.^79–81 Additional sequelae of ECI not captured in this study, such as chronic pain or physical disability, may additionally contribute to greater prevalence of SI. Future work must evaluate these potential risk factors of polytrauma related SI to better understand the long-term consequences of ECI and its relationship with post-TBI SI.

Traumatic brain injury with polytrauma and resulting depression and suicide affect not only the civilian population but also the military populations.^82,83 Future work may focus on the overlap and differences in causal mediators of suicidality identified among civilians with TBI in this report and veteran populations. Veterans experiencing TBI and polytrauma present a unique patient population, experiencing different comorbidities and psychiatric outcomes post-TBI compared with civilian populations, potentially requiring unique methods of SI screening and treatment.⁸⁴ Finally, inflammatory cascades are central to TBI and ECI in addition to depression, impulsivity, and suicidality in both the general population and those with TBI. The addition of inflammatory mediators as biological factors may further support risk model prediction for SI.^85–87

Our results build upon previous work from our group identifying ECI’s relationship to SI as well as employment and substance misuse following moderate to severe TBI.^30,35 Extracranial injury presence and severity are related to acute care complications and mortality as well as poor, long-term prognosis of global and functional recovery.^88,89 Our present study adds to a growing body of research that highlights the necessity of considering acute, nonneurological risk factors in not only mortality but also survivor-based outcomes in moderate to severe TBI.^{30,35,42,90–92}

Limitations

Inclusion into this study required year 1 and/or year 2 follow-up post-TBI, which may result in selection bias,⁹³ possibly excluding individuals at greater risk for adverse outcome. Excluding students and adults 60 years of age and older may introduce sample bias and limit generalizability regarding older adults who remain in the workforce. Variables related to preinjury suicide attempt and mental health history were not included in our main analyses because of data missingness but are important factors to consider. For example, Simpson and Tate⁹⁴ found a latency period between the most recent suicide attempt and current SI, which reflects the clinical importance of having such predictors. While the use of item 9 of the PHQ-9 has been validated to assess SI, this question does not assess all aspects of suicidality (eg, an actionable plan). Also, segmenting the PHQ-9 to look at concurrent SI may reduce the number of individuals who might otherwise be identified as depressed. This study did not examine suicide attempt data and the risk factors for SI may differ from those for suicide attempts and completed suicide, as previously demonstrated.³⁵ Finally, SI may be associated with other, unknown factors not collected by the TBIMS or NTDB, including posttraumatic stress disorder, behavioral dysfunction (eg, impulsivity, anger), and polypharmacy.

CONCLUSION

This study investigated the relationship between ECI and SI after TBI. Using a CLSEM approach, we identified unique temporal relationships among employment, substance misuse, and depression, and how these factors and ECI contribute to SI post-TBI. Extracranial injury remains uniquely associated with year 1 SI following TBI. Individuals with TBI and concurrent polytrauma should be monitored for SI, even without endorsing depression. Additional risk factors beyond employment, depression, and substance misuse, such as disinhibition, impaired executive function, and inflammatory profiles, must be examined as explanatory factors in the ECI-SI relationship. Screening and intervention tools can be tailored for effective risk mitigation and treatment. Risk model development based on these reported and additional factors may support depression and suicidality screening and management post-TBI.

Table 3:

Descriptive Analyses of the Relationship Between Pre-Injury Mental Health and Post-Injury Suicidal Endorsement

Variable	No SE, n (%)	SE, n (%)	P-value
Cause of injury
Violent/Gunshot	101 (84.87%)	18 (15.13%)	0.4443*
Non-violent	892 (87.80%)	124 (12.20%)
Suicide Attempt Ever
Yes	25 (64.10%)	14 (35.90%)	0.0001†
No	831 (88.40%)	109 (11.60%)
Suicide Attempt Year Before Injury
Yes	7 (53.85%)	6 (46.15%)	0.3072†
No	20 (71.43%)	8 (28.57%)
Received Mental Health Treatment Ever
Yes	148 (77.08%)	44 (22.92%)	<0.0001*
No	708 (89.85%)	80 (10.15%)
Received Mental Health Treatment Year Before Injury
Yes	73 (71.57%)	29 (28.43%)	0.09243*
No	73 (82.95%)	15 (17.05%)
Psychiatric Hospitalizations Ever
Yes	40 (74.07%)	14 (25.93%)	0.0050*
No	816 (88.12%)	110 (11.88%)
Psychiatric Hospitalizations Year Before Injury
Yes	6 (46.15%)	7 (53.85%)	0.02669†
No	36 (81.82%)	8 (18.18%)

Chi-square test:

^*,Fisher’s exact test:

^†Row percentages reported