Mortality Secondary to Unintentional Poisoning after Inpatient Rehabilitation among Individuals with Moderate to Severe Traumatic Brain Injury

Abstract

Studies have shown reduced life expectancy following moderate-severe traumatic brain injury (TBI) with death from unintentional poisoning (UP) being 11 times higher following TBI than in the general population. The characteristics of those who die of unintentional poisoning are compared with the characteristics of those who die of other causes (OC) in a retrospective cohort who received inpatient rehabilitation following TBI and enrolled in the TBI Model Systems National Database between 1989 and 2017 (n = 15,835 cases with 2,238 deaths recorded). Seventy-eight cases (3.5%) of deaths were the result of UP, 76% were the result of OC, and 20.5% were from an unknown cause. Among the UP deaths, 90% involved drugs (of these, 67% involved narcotic drugs and 14% involved psychostimulants), and 8% involved alcohol. Age-adjusted risk for UP death was associated with: white/non-Hispanic race/ethnicity, living alone, non-institutionalization, pre- and post-injury illicit drug use and alcohol/drug problem use, any alcohol use at last follow-up, better Functional Independence Measure^TM (FIM) scores, history of arrest, moderate disability (vs. severe disability or good recovery), less supervision needed, and greater anxiety. Adults who receive inpatient rehabilitation for TBI who die from UP are distinguishable from those who die of OC. Factors such as pre-injury substance use in the context of functional independence may be regarded as targets for prevention and/or intervention to reduce substance use and substance-related mortality among survivors of moderate-severe TBI. The current findings may have implications for medical care, surveillance, prevention, and health promotion.

Introduction

Individuals with traumatic brain injury (TBI) have been found to be at greater risk of death from an unintentional poisoning (UP) than individuals in the general population.¹ A study of the TBI Model Systems (TBIMS) National Database (NDB), weighted to represent those completing inpatient rehabilitation for TBI in the United States between 2001 and 2010, indicated that individuals with TBI were 2.23 times more likely to die at varied time points post-injury, and 10.68 times more likely to die of UP (identified with International Classification of Diseases [ICD]-9/ICD-10 as UP with codes of E850-E869/X40-X49 respectively) during this same time period than individuals in the general population of similar age, sex, and race.¹

Drug overdose deaths have increased significantly in the United States in recent years. In 2017, there were 70,237 fatal drug overdoses; 67.8% (47,600) involved an opioid.² An estimated 16,900,000 people in the United States reported past-year misuse of prescription psychotherapeutic drugs in 2018, which included prescription pain relievers, stimulants, and tranquilizers or sedatives^.3 From 2016 to 2017, death rates from all opioids increased, with increases driven by synthetic opioids.² The drugs most frequently involved in overdose deaths include opioids (heroin, oxycodone, methadone, morphine, hydrocodone, and fentanyl), benzodiazepines (alprazolam and diazepam), and stimulants (cocaine and methamphetamine).⁴

Cuthbert and colleagues⁵ estimated that 23% of adolescents and adults in the United States receiving acute rehabilitation for a primary diagnosis of TBI had misused alcohol in the month prior to their injury, and that 12% had used illicit drugs in the year prior to their injury. National estimates among those alive 5 years after TBI indicated that 17% of persons in the United States who received rehabilitation for TBI misused alcohol after their injury and 12% used illicit drugs.⁶ The association between TBI and substance misuse is also supported by research examining the client population receiving treatment for substance use disorders. Although it is estimated that one in five adults in the general population have had at least one TBI with loss of consciousness sometime in their lives,^7,8 the prevalence among persons being treated for substance use disorders may be as high as 50%^9,10 and even higher (70–80%) among persons dually diagnosed with severe mental illness and substance use disorders.^11,12

Pain, both acutely and chronically from multiple sources, is a common complication of TBI and the associated trauma, including, for example, headache and pain from orthopedic and nerve injuries.¹³ In a 2008 systematic review¹⁴ that included 4206 patients with mild to severe TBI enrolled in 23 studies, 966 (57.8%) reported chronic headache. This review found that severity of TBI was significantly associated with prevalence, with 75.3% of patients with mild TBI and 32.1% of patients with moderate and severe TBI reporting chronic headache. Further, the level of disability associated with post-traumatic headache was considerable, with average pain ratings of 5.7 (on a 0–10 scale), even for patients 5 years post-injury.¹⁵ Other pain syndromes following TBI included complex regional pain syndrome (13%), heterotopic ossification (11%), and peripheral neuropathies (10%). Also, patients may have pain from concomitant injuries, which increases the likelihood that patients may be prescribed opioids.¹⁶ For example, in a study of 2130 TBI patients receiving acute rehabilitation (enrolled between October 2008 and September 2011), 72% were prescribed a narcotic analgesic during their stay.¹⁶

Little is known about deaths from UP following TBI. The large, longitudinal TBIMS NDB provides the opportunity to learn about UP deaths among individuals who received inpatient rehabilitation for a primary diagnosis of TBI. This study explores the prevalence of UP post-TBI and the characteristics of those who die after TBI from UP compared to other causes (OC). We hypothesized UP death, versus OC deaths, would be associated for many participant characteristics (e.g., substance use) and post-injury function (e.g., substance use, physical, cognitive, behavioral, emotional, and social function), even after controlling for age of injury.

Methods

Participants and data source

The TBIMS NDB funded by the National Institute on Disability, Independent Living, and Rehabilitation Research (NIDILRR) served as the data source for this study. The TBIMS NDB includes longitudinal data on individuals with TBI receiving inpatient rehabilitation at 20 TBIMS centers dating from 1987 to the current year.¹⁷ The TBIMS NDB defines TBI as “damage to brain tissue caused by an external mechanical force as evidenced by medically documented loss of consciousness or post-traumatic amnesia (PTA) due to brain trauma or by objective neurological findings that can be reasonably attributed to TBI on physical examination or mental status examination.”^18,19 Additional eligibility criteria include (1) a diagnosis of moderate to severe TBI as determined by one of the following: PTA >24 h, trauma-related intracranial neuroimaging abnormalities, loss of consciousness >30 min, a Glasgow Coma Scale (GCS) score in the emergency department of <13 (unless caused by intubation, sedation, or intoxication); (2) being ≥16 years of age at the time of injury; (3) being admitted to a model system center's acute care hospital within 72 h of injury; (4) both acute hospital care and comprehensive rehabilitation received in a designated brain injury program within a model system center; and 5) providing informed consent to participate or having a proxy consent.^18,19 Human subject research approval is in place from each center's respective institutional review boards. Standard operating procedures dictate the methods for enrollment, follow-up, data collection, and data management.^17,19 Medical and administrative records, standardized patient assessments, and patient/family interview provide the source for demographic, pre-morbid injury, and treatment-related data. Follow-up data collection occurs by telephone or in person using structured interviews. Data collectors undergo formalized training and intermittent inter-reliability testing. Test–retest reliability of the TBIMS NDB follow-up interview is good to excellent.²⁰

Procedures

Patients were recruited during their inpatient rehabilitation hospitalization for TBI. Data were collected from a combination of medical record documentation and questionnaires during the inpatient rehabilitation stay, with follow-up telephone interviews conducted at 1, 2, and 5 years post-injury, and every subsequent 5-year interval. For this study, data from the last successful follow-up before death were utilized. Deaths were identified during routine telephone follow-up and, adhering to the TBIMS protocol for loss to follow-up, obituary and Social Security searches were conducted. When a participant was identified as having died, an official death certificate was requested from the state or the family.

Cause of death coding

From 1989 to present, using the causes of death listed on the actual death certificates, codes are assigned according to the National Data and Statistical Center (NDSC) procedures using ICD-9 codes. UP cases were identified by the primary cause of death reported on the death certificate within the ICD-9 code range of E850–E869 (UP by drugs, medicinal substances, biologicals; and other solid and liquid substances, gases, and vapors) as defined in Table 1.

Table 1.

ICD-9 Coding for Unintentional Poisoning (n = 78)

ICD-9 code	Number of deaths	Unintentional poisoning specific causes of death by ICD-9 codes
850.0	14	Opiates and related narcotics – Heroin
850.1	6	Opiates and related narcotics – Methadone
850.2	14	Opiates and related narcotics – Other (codeine, meperidine, morphine)
850.4	1	Aromatic analgesics
850.6	1	Unspecified analgesics and antipyretics
850.8	1	Other specified analgesics and antipyretics
852.8	1	Other specified sedatives and hypnotics
854.2	4	Psychostimulants
854.3	3	Central nervous system stimulants
855.2	2	Local anesthetics
858.2	1	Agents affecting blood constituents
858.6	1	Agents acting on muscles and respiratory system
858.8	2	Other specified drugs
858.9	19	Unspecified drugs
860.0	2	Alcohol
860.1	1	Other/unspecified ethyl alcohol and its products
860.9	3	Unspecified alcohol
869.1	1	Sulfur dioxide
869.8	1	Other specified gases and vapors

ICD, International Classification of Diseases.

Measures

Predictor variables selected for analysis were categorized into four domains: (1) socio-demographics, (2) injury severity, (3) alcohol and illicit drug use, and (4) function, community participation, mood, and life satisfaction. These variables were assessed at time of injury, inpatient rehabilitation discharge, and/or last available post-injury follow-up. Specific groupings for the categorical predictors variables are delineated in Tables 2–4 with selected relevant details described in the following sections.

Table 2.

Participant Characteristics by Cause of Death Group

		Cause of death
Variable	Group	Unintentional poison (n = 78)	Other cause (n = 1708)	p value
Age group at injury (years)	16–29	27 (34.6%)	115 (6.7%)		< 0.001
	30–39	22 (28.2%)	125 (7.3%)
	40–49	20 (25.6%)	248 (14.5%)
	50–59	6 (7.7%)	305 (17.9%)
	60–69	3 (3.8%)	275 (16.1%)
	70–79	0 (0.0%)	328 (19.2%)
	80 or older	0 (0.0%)	312 (18.3%)
Age at injury (years)	Median (IQR)	33.5 (25.0, 42.0)	62.0 (48.0, 76.0)	< 0.001
Age at death (years)	Median (IQR)	40.7 (30.9, 49.2)	68.0 (53.6, 81.0)	< 0.001
Death yeara	1991–1995	0 (0.0%)	32 (1.9%)	0.603
	1996–2000	5 (6.4%)	86 (5.0%)
	2001–2005	14 (17.9%)	240 (14.1%)
	2006–2010	24 (30.8%)	578 (33.9%)
	2011–2015	34 (43.6%)	698 (40.9%)
	2016–2017	1 (1.3%)	71 (4.2%)
	[missing]	[0]	[3]
Sex	Male	62 (79.5%)	1227 (71.9%)	0.143
	Female	16 (20.5%)	480 (28.1%)
	[missing]	[0]	[1]
Race/Ethnicity	White/Non-Hispanic	63 (80.8%)	1253 (73.4%)	0.146
	All other	15 (19.2%)	455 (26.6%)
Pre-injury marital status				< 0.001
	Never married	42 (53.8%)	346 (20.3%)
	Married	11 (14.1%)	704 (41.3%)
	Previously married	25 (32.1%)	655 (38.4%)
	[missing]	[0]	[3]
Pre-injury employment status				< 0.001
	Employed	36 (46.2%)	540 (31.9%)
	Unemployed	23 (29.5%)	234 (13.8%)
	Retired	7 (9.0%)	792 (46.8%)
	Other	12 (15.4%)	128 (7.6%)
	[missing]	[0]	[14]
Pre-injury drug use	Yes	37 (51.4%)	212 (13.8%)	< 0.001
(n = 1613; 173 missing, 10%)	No	35 (48.6%)	1329 (86.2%)
	[missing]	[6]	[167]
Pre-injury drinking category	Abstaining	18 (29.5%)	714 (55.6%)	0.005
(n = 1345; 441 missing, 25%)	Light	8 (13.1%)	143 (11.1%)
	Medium	15 (24.6%)	197 (15.3%)
	Heavy	20 (32.8%)	230 (17.9%)
	[missing]	[17]	[424]
Cause of Injury	Falls	15 (19.2%)	864 (50.7%)	< 0.001
	Vehicular	42 (53.8%)	473 (27.7%)
	Violence	9 (11.5%)	220 (12.9%)
	Other	12 (15.4%)	148 (8.7%)
	[missing]	[0]	[3]
Glasgow Coma Scale	Mild	25 (32.5%)	927 (55.5%)	< 0.001
	Moderate	9 (11.7%)	210 (12.6%)
	Severe	24 (31.2%)	348 (20.9%)
	Sedated	19 (24.7%)	184 (11.0%)
	[missing]	[1]	[39]

^aOnly computed for subjects with known death date.

IQR, interquartile range.

Table 3.

Age^a Adjusted Odds Ratios of Unintentional Poisoning versus Other Cause of Death for Pre-Injury and Injury Characteristics

Predictor variable	Comparison	Global p value	OR	95% CI	Pairwise p value
Sex	Male vs. Female	0.9813	1.01	(0.56, 1.82)
Race/Ethnicity	White/Non-Hispanic vs. All other	0.0054*	2.34	(1.29, 4.27)
Pre-injury marital status		0.0530
	Never married vs. Married		1.45	(0.68, 3.14)	0.3373
	Previously married vs. Married		2.40	(1.15, 5.01)	0.0201
	Previously married vs. Never Married		1.65	(0.87, 3.12)	0.1266
Pre-injury living with		0.2119
	Alone vs. Spouse		2.38	(1.05, 5.39)	0.0377
	Parents vs. Spouse		1.92	(0.79, 4.67)	0.1487
	Other vs. Spouse		2.06	(0.91, 4.63)	0.0816
	Alone vs. Parents		1.24	(0.61, 2.53)	0.5582
	Other vs. Parents		1.07	(0.56, 2.05)	0.8392
	Alone vs. Other		1.16	(0.61, 2.20)	0.6541
Discharge living with		0.1063
	Alone vs. Spouse		2.29	(0.66, 8.01)	0.1937
	Parents vs. Spouse		2.41	(1.01, 5.73)	0.0469
	Other vs. Spouse		1.32	(0.58, 2.99)	0.5041
	Alone vs. Parents		0.95	(0.30, 2.99)	0.9329
	Other vs. Parents		0.55	(0.30, 0.99)	0.0468
	Alone vs. Other		1.74	(0.57, 5.23)	0.3286
Pre-injury residence	Not private vs. Private	0.6090	1.33	(0.45, 3.90)
Discharge residence	Not private vs. Private	0.3008	0.71	(0.37, 1.36)
Cause of injury		0.3129
	Falls vs. Violence		1.56	(0.64, 3.78)	0.3254
	Vehicular vs. Violence		1.98	(0.92, 4.24)	0.0800
	Other vs. Violence		2.13	(0.84, 5.37)	0.1007
	Vehicular vs. Falls		1.27	(0.64, 2.50)	0.4927
	Other vs. Falls		1.36	(0.58, 3.19)	0.4737
	Other vs. Vehicular		1.08	(0.53, 2.19)	0.8415
Rehabilitation payer source		0.4723
	Government vs. Private		1.38	(0.82, 2.32)	0.2229
	Government vs. Other		1.22	(0.48, 3.11)	0.6842
	Other vs. Private		1.14	(0.44, 2.97)	0.7942
Pre-injury employment status		0.4821
	Unemployed vs. Employed		1.45	(0.82, 2.56)	0.2002
	Retired vs. Employed		0.89	(0.35, 2.25)	0.7988
	Other vs. Employed		1.45	(0.69, 3.04)	0.3311
	Unemployed vs. Retired		1.64	(0.63, 4.27)	0.3131
	Other vs. Retired		1.63	(0.54, 4.97)	0.3880
	Other vs. Unemployed		1.00	(0.45, 2.21)	0.9931
Pre-injury drug use	Yes vs. No	0.0053*	2.12	(1.25, 3.60)
Pre-injury alcohol use	Yes vs. No	0.2793	1.39	(0.77, 2.52)
Pre-injury drinking category		0.3874
	Heavy vs. Abstaining		1.76	(0.87, 3.54)	0.1145
	Medium vs. Abstaining		1.57	(0.73, 3.35)	0.2444
	Light vs. Abstaining		1.10	(0.44, 2.78)	0.8352
	Heavy vs. Light		1.59	(0.64, 3.98)	0.3186
	Medium vs. Light		1.42	(0.55, 3.71)	0.4702
	Heavy vs. Medium		1.12	(0.53, 2.37)	0.7672
Pre-injury alcohol/drug problem	Yes vs. No	0.0355*	1.86	(1.04, 3.31)
Glasgow Coma Scale		0.1952
	Mild vs. Severe		1.90	(0.98, 3.71)	0.0583
	Moderate vs. Severe		0.84	(0.36, 1.92)	0.6742
	Sedated vs. Severe		1.76	(0.90, 3.43)	0.0989
	Mild vs. Moderate		1.59	(0.70, 3.65)	0.2709
	Sedated vs. Moderate		1.47	(0.62, 3.49)	0.3826
	Mild vs. Sedated		1.08	(0.54, 2.18)	0.8226
Years from discharge to death	1 year decrease	0.1119	1.04	(0.99, 1.10)
DRS at discharge	1 unit decrease	0.0014*	1.14	(1.05, 1.24)
FIM™ Motor at discharge	1 unit increase	0.0029*	1.02	(1.01, 1.04)
FIM Cognitive at discharge	1 unit increase	0.0009*	1.07	(1.03, 1.11)
FIM Total at discharge	1 unit increase	0.0012*	1.02	(1.01, 1.03)

^*Age-adjusted global test for predictor variable statistically significant at α = 0.05.

^aAge, age at the time of injury

OR, odds ratio; CI, confidence interval; DRS, Disability Rating Scale; FIM, Functional Independence Measure.

Table 4.

Age-Adjusted Odds Ratios of Unintentional Poisoning versus Other Cause of Death for Post-Injury Characteristics Measured at Last Follow-Up

Predictor variable at last FU	Comparison	Global p value	OR	95% CI	Pairwise p value
Marital status		0.1247
	Not married vs. Married		2.05	(0.82, 5.09)	0.1243
	Previously married vs. Married		2.49	(1.04, 5.97)	0.0415
	Previously married vs. Not married		1.22	(0.62, 2.37)	0.5677
Living with		0.0108*
	Alone vs. Spouse		3.58	(1.44, 8.90)	0.0059	††
	Parents vs. Spouse		2.10	(0.83, 5.31)	0.1151
	Other vs. Spouse		1.43	(0.60, 3.40)	0.4207
	Alone vs. Parents		1.70	(0.84, 3.45)	0.1159
	Other vs. Parents		0.68	(0.36, 1.26)	0.2217
	Alone vs. Other		2.51	(1.32, 4.77)	0.0049	††
Residence	Private vs. Not private	0.0457*	2.29	(1.02, 5.14)
Community type		0.3435
	Urban vs. Rural		1.35	(0.68, 2.67)	0.3889
	Rural vs. Suburban		1.25	(0.56, 2.78)	0.5876
	Urban vs. Suburban		1.68	(0.82, 3.48)	0.1588
Employment status		0.1014
	Unemployed vs. Employed		2.58	(1.03, 6.48)	0.0431
	Retired vs. Employed		1.99	(0.79, 5.05)	0.1456
	Employed vs. Other		2.27	(0.26, 19.96)	0.4614
	Unemployed vs. Retired		1.30	(0.70, 2.39)	0.4046
	Retired vs. Other		4.52	(0.57, 36.02)	0.1547
	Unemployed vs. Other		5.85	(0.74, 46.60)	0.0950
Drug use	Yes vs. No	<0.0001*	3.72	(1.96, 7.03)
Alcohol use	Yes vs. No	0.0012*	2.62	(1.47, 4.69)
Drinking category		0.0074*
	Heavy vs. Abstaining		2.88	(1.20, 6.89)	0.0177	†
	Medium vs. Abstaining		3.05	(1.42, 6.54)	0.0042	††
	Light vs. Abstaining		2.49	(1.14, 5.45)	0.0225	†
	Heavy vs. Light		1.16	(0.43, 3.11)	0.7749
	Medium vs. Light		1.22	(0.50, 3.00)	0.6586
	Medium vs. Heavy		1.06	(0.40, 2.81)	0.9079
Alcohol/Drug problem	Yes vs. No	0.0002*	3.07	(1.72, 5.48)
Transportation	Does not drive vs. Drives	0.3958	1.15	(0.83, 1.59)
Arrests	Yes vs. No	0.0002*	4.05	(1.96, 8.38)
GOSE		0.0002*
	Good recovery vs. Severe disability		1.07	(0.41, 2.83)	0.8859
	Moderate disability vs. Severe disability		3.60	(1.83, 7.09)	0.0002	†
	Moderate disability vs. Good recovery		3.35	(1.41, 7.96)	0.0061	†
SRS		0.0063*
	Overnight/Part-time vs. Full-time		5.52	(1.26, 24.20)	0.0234	†
	Independent vs. Overnight/Part-time		1.63	(0.91, 2.92)	0.1034
	Independent vs. Full-time		8.98	(2.10, 38.41)	0.0031	††
DRS	1 unit decrease	0.0837	1.05	(0.99, 1.10)
FIM™ Motor	1 unit increase	0.0019*	1.03	(1.01, 1.05)
FIM Cognitive	1 unit increase	0.0007*	1.08	(1.03, 1.13)
FIM Total	1 unit increase	0.0010*	1.03	(1.01, 1.04)
SWLS	1 unit decrease	0.3388	1.02	(0.98, 1.06)
PHQ-9	1 unit increase	0.1070	1.06	(0.99, 1.13)
GAD 7	1 unit increase	0.0349*	1.11	(1.01, 1.23)
PART-O Productivity	1 unit decrease	0.7267	0.91	(0.53, 1.56)
PART-O Out and About	1 unit decrease	0.0636	1.60	(0.97, 2.63)
PART-O Social	1 unit decrease	0.5223	1.13	(0.78, 1.65)

^*Age-adjusted (age at the time of injury) global test for predictor variable statistically significant at α = 0.05; †Global test statistically significant at α = 0.05 for categorical predictors with more than two categories and associated post-hoc pairwise comparisons statistically significant at α = 0.05; ††Relevant pairwise comparison statistically significant after Bonferroni adjustment for multiple comparisons (3 categories α = 0.0167; 4 categories α = 0.0083).

OR, odds ratio; CI, confidence interval; FU, follow-up; GOSE, Glasgow Coma Scale – Extended; SRS, Supervision Rating Scale; DRS, Disability Rating Scale; FIM, Functional Independence Measure; SWLS, Satisfaction with Life Scale; PHQ, Patient Health Questionnaire (9-item); GAD, General Anxiety Disorder Assessment (7-item); PART-O, Participation Assessment with Recombined Tools-Objective (17-item).

Socio-demographic characteristics

These variables included age at injury and death, year of death, sex, race/ethnicity, marital status, living situation, residence, employment status, primary rehabilitation payer source, geographical region/community type, transportation use, and recent arrest history.

TBI characteristics

TBI characteristics included cause of injury, GCS score at the time of emergency department admission, and years from rehabilitation discharge to death.

Alcohol and illicit drug use

Data on alcohol use in the past month and illicit or non-prescription drug use in the past year are collected for the time periods prior to injury and the last follow-up. Drinking categories were derived from the number of drinks per week using the Centers for Disease Control and Prevention's Behavioral Risk Factor Surveillance System User Guide.²¹ Drinking categories were defined as “Abstaining,” “Light” (≤ 3 drinks per week), “Medium” (3–7 drinks per week, women; 3–14 drinks per week, men), “Heavy” (>7 drinks per week, women; >14 drinks per week, men), or “binge drinking” (4 or more drinks for women or 5 or more drinks for men on one occasion). Risk assessments were based on the United States Department of Health and Human Services Dietary Guidelines.²² Alcohol/drug problem was defined as any illicit drug use (past year) or heavy alcohol consumption or binge drinking (past month).¹⁹

Function, community participation, mood, and life satisfaction

Functional measures were contained within the Functional Independence Measure™ (FIM) and Disability Rating Scale (DRS), at rehabilitation discharge and last follow-up, as well as the Supervision Rating Scale (SRS) at last follow-up. The FIM measures functional independence²³ using 18 items scored from 1 (needing total assistance) to 7 (complete independence), with higher scores indicating greater functional independence. FIM has two subscales: the 13-item Motor subscale (scores range from 13 to 91) and the 5-item Cognitive Scale (scores range from 5 to 35). The eight-item DRS assesses cognitive ability to manage activities of daily living, need for assistance or supervision, and employability.²⁴ DRS scores range from 0 (no disability) to 29 (extreme vegetative state). The SRS is a 13-point ordinal scale that describes the level of supervision received by the individual with TBI.²⁵ SRS ratings were trichotomized to 1–2 (independent), 3–7 (needing overnight or part-time supervision), and 8–13 (needing full-time direct or indirect supervision).

Community participation was represented by transportation access, the Glasgow Outcome Scale – Extended (GOS-E), and the Participation Assessment with Recombined Tools (PART-O) scores from the last follow-up. The GOS-E was used as a seven-point scale with the ability to live, work, travel, and socialize independently measured on a scale of 2 (vegetative state) through 8 (upper-good recovery).^26–29 The Participation Assessment with Recombined Tools-Objective (PART) is a 17-item measure of participation with three domains: Productivity, Being Out and About, and Social Relations, each ranging from 0 to 5, with higher scores indicative of greater community participation.³⁰

Life satisfaction, depression, and anxiety at last follow-up were assessed through self-report using the Satisfaction with Life Scale (SWLS), Patient Health Questionnaire 9 (PHQ-9), and Generalized Anxiety Disorder Assessment (GAD-7). The SWLS³¹ is a five-item global measure of life satisfaction³¹ rated on a seven-point Likert scale; total scores range between 5 and 35, with higher scores indicative of greater life satisfaction. The PHQ-9³² uses a three-point Likert scale to rate the frequency of specified problems of depression during the past 2 weeks. Total scores range between 0 and 27, with higher scores indicating more depressive symptoms.^32,33 The GAD-7³⁴ measures the frequency of seven symptoms from the Diagnostic and Statistical Manual of Mental Disorders, 4th edition (DSM-IV) criteria for generalized anxiety disorder. Total scores range between 0 and 21, with higher scores indicating more anxiety symptoms.

Statistical analyses

All analyses were conducted using SAS v.9.4³⁵; results were considered significant if p < 0.05, unless otherwise stated. Select participant characteristics were summarized by group using frequency counts and percentages for nominal characteristics. Because most continuous characteristics exhibited a fair degree of skewness, medians and interquartile ranges (IQRs) were chosen to describe the center and spread of these variables, respectively. The characteristics were compared between UP and OC groups using non-parametric χ² tests (categorical characteristics) and Wilcoxon rank-sum tests (continuous characteristics).

Our preliminary analyses revealed a substantial difference in age at injury between UP and OC death groups (median age 33.5 vs. 62.0 years, respectively); therefore, all subsequent analyses controlled for this variable. Specifically, we estimated the age-adjusted relationship between each predictor variable and outcome (UP vs. OC death) using multivariable logistical regression methods by fitting separate models for each predictor variable solely adjusted for age at injury (as a continuous variable). These relationships were quantified with age-adjusted odds ratios (ORs) of UP versus OC death from these models along with 95% confidence intervals (CIs) expressing the level of uncertainty in these estimates. For categorical predictors with more than two levels, we conducted post-hoc analyses to estimate the pairwise ORs among the varying levels of the predictor, in order to further clarify the relationship between these predictors and UP versus OC death. As is standard practice, we did not interpret pairwise post-hoc tests as significant unless the global test for the predictor variable was significant (p < 0.05). Finally, we utilized Bonferroni corrections to adjust these pairwise post-hoc tests for multiple comparisons.

Results

Participants

Figure 1 depicts the sample flow chart. There were 15,835 participants enrolled into the TBIMS NDB with rehabilitation discharge dates between March 4, 1989 and March 31, 2017, with an overall follow-up rate of 81%. We excluded those who were known to be alive at last follow-up (n = 13,581) or who died during inpatient rehabilitation (n = 16), leaving a final sample of 2238 who died after rehabilitation discharge.

FIG 1.

Sample flow chart.

Seventy-eight of the 2238 (3.5%) died from UP, whereas 1708 (76%) died from OC, and the remaining 452 (20.5%) died from unknown causes. Table 1 describes the number of UP deaths by specific ICD-9 code. Among the 78 UP deaths, 70 (90%) involved drugs and 6 (8%) involved alcohol. Of the 70 UP deaths involving drugs, 19 (27%) involved an unspecified drug. Of the 51 deaths involving a specified drug, 34 (67%) involved narcotic drugs and 7 (14%) involved psychostimulants. The median time from injury to death from UP for these 78 participants was 4.4 years (IQR = 1.6–8.0), with 16 (21%) deaths occurring in the 1st year post-injury, 12 (15%; 36% cumulative) in the 2nd year, and another 23 (29%; 65% cumulative) between years 2 and 5 post-injury. Of the 1708 OC deaths, the median time to death was 3.6 years (IQR = 1.4–7.3). Within the 1st year following TBI, 439 (26%) died, with 280 (16%; 42% cumulative) deaths between year 1 and year 2, and 444 (26%; 68% cumulative) deaths between years 2 and 5 (data not shown).

Basic demographic and injury characteristics for the UP and OC deaths are summarized in Table 2. Age at time of injury (p < 0.001) and age at time of death (p < 0.001) were found to be significantly younger for the UP group than for the OC group. Significant group differences were also found for pre-injury marital status, employment status, drug use, drinking category, cause of injury, and GCS. Year of death was not statically different between the UP and OC groups (p = 0.60). As most of these group differences (with the exception of GCS) are expected as a result of differences in age, we conducted all subsequent analyses adjusting age at injury.

Analyses of participant characteristics adjusted for age at injury

We estimated the age-adjusted relationships between each predictor variable and the odds of death from UP versus OC. These relationships are summarized in Table 3 for pre-injury and injury characteristics and in Table 4 for post-injury characteristics. After controlling for age at injury, the following characteristics were found to be associated with an increased odds of UP death versus OC death (global p value <0.05): race/ethnicity, who they are living with at last follow-up, type of residence at last follow-up, drug use in the year before injury and before last follow-up, alcohol use and drinking category in the month before last follow-up, drug/alcohol problem pre-injury and at the last follow-up, arrests in the year prior to last follow-up, GOS-E at last follow-up, SRS at last follow-up, DRS at discharge, FIM Motor, FIM Cognitive, and FIM Total at both discharge and at last follow-up, and GAD-7 at last follow-up. The odds of UP death versus OC death were significantly greater for: white/non-Hispanic versus all other (2.34); living alone versus with spouse (3.58) or other (2.51) at last follow-up; living in private versus non-private residence at last follow-up (2.29); illicit drug use pre-injury (2.12) and at last follow-up (3.72); any alcohol use at last follow-up (2.62); heavy (2.88), medium(3.05), or light (2.49) versus the abstaining from drinking category at last follow-up; alcohol/drug problem use pre-injury (1.86) and at last follow-up (3.07); arrests in the year prior to last follow-up (4.05); moderate disability versus severe disability (3.60) or good recovery (3.35) on GOS-E at last follow-up; overnight/part-time supervision (5.52) or independent (8.97) versus full-time supervision (SRS) needed at last follow-up; lower (better) DRS total scores at rehabilitation discharge (1.14); higher (better) FIM Motor (1.02), Cognitive (1.07), and Total (1.02) scores at rehabilitation discharge; higher (better) FIM Motor (OR = 1.03), Cognitive (1.08), and Total (1.03) scores at last follow-up; and greater anxiety (GAD-7) at last follow-up (1.11).

Discussion

This study examined the frequency of UP deaths and characteristics of UP deaths compared with OC deaths after TBI, to further understand the risks for and circumstances of these deaths. Among individuals who received inpatient rehabilitation for TBI and died subsequent to rehabilitation discharge, 3.5% subsequently died of UP. Those who died of UP differed markedly from those who died of OC; namely, they were younger, more independent, higher functioning, reported higher levels of anxiety, and were more likely to have misused drugs and/or alcohol prior to their injury. Of those who died from UP, the vast majority of deaths involved drugs, with 67% involving narcotic drugs. These findings shed light on an important and perhaps overlooked subgroup of TBI survivors who, by all traditional metrics, have experienced a relatively good outcome after serious head trauma. Compared with those who achieve lower levels of functional recovery and who continue to require supervision for years after injury, those who are younger and more independent may be at risk for substance misuse and its devastating consequences. Individuals in the current study sustained a TBI of sufficient severity to necessitate intensive inpatient rehabilitation, and many such individuals experience lifelong cognitive impairments, however subtle. TBI-related neurobehavioral changes such as executive dysfunction, poor impulse control, and mood disturbance may confer greater risk for substance misuse and use disorder. In addition, greater functional independence and lower levels of supervision are factors that increase access to illicit and/or prescription drugs. Findings from the current study underscore the importance of ongoing monitoring for these individuals, particularly those who have past histories of substance misuse, prior arrests, and anxiety.

Clinical implications

Recently, opioid misuse was recognized as a public health emergency,³⁶ with subgroups of individuals at elevated risk for opioid use disorder and its most dire consequences. Opioid exposure may be of particular concern following TBI. Our prior work using a TBIMS data set found that individuals with TBI were at 11 times greater risk for UP death than those in the general population of similar age, sex, and race.¹ The present findings illustrate that these post-TBI UP deaths commonly involved opioids and psychostimulants. Individuals with TBI report high rates of chronic pain,^14,37,38 and pain can interfere with engagement in rehabilitation therapies and progress. Given this, patients may be prescribed opioids.¹⁶

In 2016 and 2018, respectively, the Centers for Disease Control and Prevention (CDC) developed a guideline intended to improve communication between clinicians and patients about the risks and benefits of opioid therapy for chronic pain, and the American Academy of Physical Medicine and Rehabilitation (AAPMR) developed a guideline that recommends using opioids only when the benefits outweigh the risks.^39,40 The Veteran's Association and Department of Defense clinical practice guidelines specifically caution against the use of opioid medications in the context of TBI,⁴¹ but a recent study suggests that these guidelines are not routinely followed.⁴² The AAPMR guidelines also recommend restricting opioid use to ≤14 days for acute pain, with regular reassessment as well as multimodal pain treatment, including psychological interventions, exercise, and interventional procedures. Ideally, opioid prescription is avoided for individuals with TBI, or only used briefly and with close monitoring.¹⁶

For the patient with TBI, our findings suggest that important considerations for risk stratification include pre-TBI substance use history; degree of neurobehavioral impairment including awareness, mood, and behavioral self-regulation; and level of anxiety and pain; along with the other risk factors noted. Inpatient screening for these risk factors can easily be accomplished with several evidence-based instruments. For pre-TBI substance abuse, the 12-item form of the Screener and Opioid Assessment for Patients with Pain-Revised⁴³ is well validated, quick, and easy to administer. There is no short, standardized screening tool for relevant neurobehavioral disorders for use in an inpatient rehabilitation setting, but certain items from the Neurobehavioral Rating Scale-Revised⁴⁴ may be useful. For example, this instrument includes items for anxiety, disinhibition, and lability of mood that are based on clinician rating of either absent, mild, moderate, or severe impairment. These factors can be considered when making medication prescription decisions prior to discharge from the structure of a hospital setting and during follow up assessments. One tool that can enhance effective prescribing decisions is routine use of the Prescription Drug Monitoring Programs (PDMP) established by states. This study highlights the importance of PDMP use by clinicians to maximize understanding of a patient's complete opioid use history⁴⁵ from possible multiple sources. Receiving opioids from multiple prescribers or pharmacies is associated with greater likelihood of subsequent medically assisted treatment prescription, such as buprenorphine.⁴⁶ It is particularly important to communicate a plan to the patient, family, and other providers for alternatives to opioid use, opioid discontinuance, or management and follow-up. It is also especially important to document the relevance of the TBI as a risk factor for possible opioid misuse, as often the prescribers of these medications are unaware of the patient's history of TBI. Patient and family education on the importance of monitoring prescription use for pain and availability of resources such as psychotherapy, stress management, and cognitive rehabilitation can be a part of rehabilitation discharge instructions. The pattern of results suggests that the data collected at last follow-up may be more predictive of UP compared with OC death than data collected earlier. This speaks to the need for ongoing monitoring for this population.

Future research directions

Additional research is needed to further explore risk factors for UP, including the factors identified in this study. For example, research is needed to understand the degree to which UP is related to opioid use for chronic pain following TBI. Research on opioid use within the context of pain management after moderate and severe TBI is a first step to this end, and is one aim of a multi-center TBIMS grant (2018-2023). The current study was not designed to distinguish factors associated with substance use from characteristics associated with overdose death secondary to substance use, and further research is required to inform a more detailed risk stratification system and identify post-TBI patients who use substances who are at highest risk for UP. Studies such as these are needed to inform medical care, surveillance, prevention, and health promotion efforts that will ultimately reduce the incidence of UP following TBI. Future research can examine the relationship of pre-injury and personal characteristics to better understand the relationship with post-injury depression and UP.

Study limitations

Because this is a study of people who received inpatient rehabilitation, it is unknown how generalizable these results may be for individuals with TBI who do not receive inpatient rehabilitation. A recent study reports a higher percentage of substance use in individuals with milder injuries seen in the emergency department, who are more likely to engage in substance use in the 1st year after TBI. Many of these individuals are not offered rehabilitation or medical follow-up following their injury and may not have received information on prevention of substance use.⁴⁷ The current analyses were limited to those variables collected by the TBIMS NDB, and as such, several possible risk factors for substance misuse and overdose were not available for inclusion in this investigation. Cause of death was unknown for 20% of known deaths, and it is possible that there may have been unknown deaths among those lost to follow-up. Death certificates may be missing or contain inaccuracies resulting from misclassification, errors, and incomplete information.^48–50 This study used patient data collected over a long period of time; meanwhile, the opioid crisis emerged within the latter half of the study period. The TBIMS NDB utilizes guidelines for coding and assigning the primary cause of death to minimize such problems, but important details such as the exact agent causing toxicity is not known from the ICD-9 and death certificate level data. It is, therefore, not known whether multiple drugs and/or alcohol were consumed in UP cases, and it is similarly unknown whether drugs were prescribed or obtained illicitly. It is important to consider that death certificates are known to underestimate UP deaths,^51–53 and it is not known if any of these deaths deemed to be caused by UP were actually intentional. It is of note that self-reported depression at last follow-up was not found to be different between those dying from UP and those dying from OC. Lastly, although the analyses were adjusted for age at injury, the relatively small sample size prevented multivariate modeling.

Conclusion

This study found differences in those with TBI who died of UP compared with those who died of OC, revealing risk factors that may aid improved surveillance, identification, treatment, and prevention. After controlling for age at injury, adults with TBI who received inpatient rehabilitation and later died of UP had better functional outcomes and were more independent than those who died of OC, although they were more likely to have substance use problems and a history of arrest. More than half of the UP deaths involving a known drug were narcotic related.

Funding Information

This research was supported by an interagency agreement between the United States Department of Health and Human Services (HHS), CDC, and the United States Department of Education, National Institute on Disability and Rehabilitation Research (NIDRR) with supplemental funding to the NIDRR-funded Traumatic Brain Injury Model Systems National Data and Statistical Center (Grant Numbers 90DP0013 and 90DP0084). In 2014, NIDRR was moved to the Administration for Community Living (ACL) within HHS, and renamed the National Institute on Disability, Independent Living, and Rehabilitation Research (NIDLRR). The contents of this publication were developed under grants from the NIDILRR: Indiana University School of Medicine/Rehabilitation Hospital of Indiana (Grant Numbers 90DP0036 and 90DRTB0002), Icahn School of Medicine at Mount Sinai (Grant Numbers 90DP0038 and 90DPTB0009), Ohio State University (Grant Number 90DP0040), and Traumatic Brain Injury Model Systems National Data and Statistical Center at Craig Hospital (Grant Numbers 90DP0013 and #90DP0084), by a grant from the National Institutes of Health, The Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD) (Grant Number K01HD074651-01A1), and by a grant from the CDC (Grant Number 5 NU17CE002721-03-00). The contents of this publication do not necessarily represent the policy of NIDILRR, ACL, CDC, or HHS, and endorsement by the federal government should not be assumed.

Disclaimer

The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the Centers for Disease Control and Prevention.

Author Disclosure Statement

No competing financial interests exist.