Using Machine Learning to Examine Suicidal Ideation After Traumatic Brain Injury: A Traumatic Brain Injury Model Systems National Database Study

Association between impulsivity, self-harm, suicidal ideation, and suicide attempts in traumatic brain injury patients

Traumatic brain injury (TBI) affects over 48 million people worldwide each year. Suicide is common in TBI, and there are several known contributing factors, including severe TBI, depression, alcohol use, and male sex. Impulsivity, or the tendency to act quickly with little thought, may be an early predictor of suicidality following TBI. The purpose of this study was to evaluate the risk of suicidality in patients with a prior history of impulsivity following a traumatic brain injury (TBI). Using de-identified electronic health records from the TriNetX United States Collaborative Network with Natural Language Processing, three cohorts were generated: the impulsivity TBI cohort (I+TBI+) included subjects with a diagnosis of impulsivity before a diagnosis of TBI; the no impulsivity TBI cohort (I-TBI+) included patients with TBI but no impulsivity; the impulsivity no TBI cohort (I+TBI-) included patients with impulsivity but TBI. Two analyses were conducted, including analysis 1 (impulsivity TBI vs. no impulsivity TBI) and analysis 2 (impulsivity TBI vs. impulsivity no TBI). Patients were 1:1 propensity score matched by age, sex, race, ethnicity, psychiatric diagnoses, and antidepressant use. Outcomes included a diagnosis of self-harm, suicidal ideation, or a suicide attempt within one year after the index event. The all-time incidence of each outcome was assessed across different age categories. The chi-square test (categorical variables) and t-test (numerical variables) were used to assess for differences between groups. A total of 1,292,776 TBI patients were identified in the study. After 1:1 propensity score-matching, there were 20,694 patients (mean [SD] age, 48.1 [21.8]; 8,424 females [40.7%]) with impulsivity and TBI (I+TBI+), 1,272,082 patients (mean [SD] age, 46.0 [23.1]; 562,705 females [44.2%]) with TBI alone (I-TBI+), and 90,669 patients (mean [SD] age, 43.7 [22.6]; 45,188 females [49.8%]) with impulsivity alone (I+TBI-). Within the first year after a TBI, patients with impulsivity were more likely to exhibit self-harm (P < 0.001), suicidal ideation (P < 0.001) or a suicide attempt (P < 0.001). Compared to TBI patients without impulsivity, those with impulsivity had a four-fold increase in the incidence of self-harm (2.81% vs. 0.63%), an eight-fold increase in suicidal ideation (52.42% vs. 6.41%), and a twenty-one-fold increase in suicide attempts (32.02% vs. 1.50%). This study suggests that impulsivity diagnosed before a TBI may increase the risk of post-traumatic suicidality, with a four-fold increased risk of self-harm, an eight-fold increased risk of suicidal ideation and a twenty-one-fold increased risk of suicide attempts. This characterizes a group of at-risk individuals who may benefit from early psychiatric support and targeted interventions following a TBI.

Suicide Attempts and Ideation Among Veterans/Service Members and Non-Veterans Over 5 Years Following Traumatic Brain Injury: A Combined NIDILRR and VA TBI Model Systems Study

OBJECTIVE

This study compared rates of suicide attempt (SA) and suicidal ideation (SI) during the first 5 years after traumatic brain injury (TBI) among veterans and service members (V/SMs) in the Veterans Affairs (VA) and the National Institute on Disability, Independent Living, and Rehabilitation Research (NIDILRR) Model Systems National Databases to each other and to non-veterans (non-Vs) in the NIDILRR database.

SETTING

Twenty-one NIDILRR and 5 VA TBI Model Systems (TBIMS) inpatient rehabilitation facilities in the United States.

PARTICIPANTS

Participants with TBI were discharged from rehabilitation alive, had a known military status recorded (either non-V or history of military service), and successful 1-, 2-, and/or 5-year follow-up interviews completed between 2009 and 2021. The year 1 cohort included 8737 unique participants (8347 with SA data and 3987 with SI data); the year 2 (7628 participants) and year 5 (4837 participants) cohorts both had similar demographic characteristics to the year 1 cohort.

DESIGN

Longitudinal design with data collected across TBIMS centers at 1, 2, and 5 years post-injury.

MAIN OUTCOMES AND MEASURES

History of SA in past year and SI in past 2 weeks assessed by the Patient Health Questionnaire-9 (PHQ-9). Patient demographics, injury characteristics, and rehabilitation outcomes were also assessed.

RESULTS

Full sample rates of SA were 1.9%, 1.5%, and 1.6%, and rates of SI were 9.6%, 10.1%, and 8.7% (respectively at years 1, 2, and 5). There were significant differences among groups based on demographic, injury-related, mental/behavioral health, and functional outcome variables. Characteristics predicting SA/SI related to mental health history, substance use, younger age, lower functional independence, and greater levels of disability.

CONCLUSIONS

Compared with participants with TBI in the NIDILRR system, higher rates of SI among V/SMs with TBI in the VA system appear associated with risk factors observed within this group, including mental/behavioral health characteristics and overall levels of disability.

Traumatic Brain Injury Rehabilitation Outcome Prediction Using Machine Learning Methods

Objective

To investigate the performance of machine learning (ML) methods for predicting outcomes from inpatient rehabilitation for subjects with TBI using a dataset with a large number of predictor variables. Our second objective was to identify top predictive features selected by the ML models for each outcome and to validate the interpretability of the models.

Design

Secondary analysis using computational modeling of relationships between patients, injury and treatment activities and 6 outcomes, applied to the large multi-site, prospective, longitudinal observational dataset collected during the traumatic brain injury inpatient rehabilitation study.

Setting

Acute inpatient rehabilitation.

Participants

1946 patients aged 14 years or older, who sustained a severe, moderate, or complicated mild TBI, and were admitted to 1 of 9 US inpatient rehabilitation sites between 2008 and 2011 (N=1946).

Main Outcome Measures

Rehabilitation length of stay, discharge to home, FIM cognitive and FIM motor at discharge and at 9-months post discharge.

Results

Advanced ML models, specifically gradient boosting tree model, performed consistently better than all other models, including classical linear regression models. Top ranked predictive features were identified for each of the 6 outcome variables. Level of effort, days to rehabilitation admission, age at rehabilitation admission, and advanced mobility activities were the most frequently top ranked predictive features. The highest-ranking predictive feature differed across the specific outcome variable.

Conclusions

Identifying patient, injury, and rehabilitation treatment variables that are predictive of better outcomes will contribute to cost-effective care delivery and guide evidence-based clinical practice. ML methods can contribute to these efforts.