Predictors for Withdrawal of Life-Sustaining Therapies in Patients With Traumatic Brain Injury: A Retrospective Trauma Quality Improvement Program Database Study

Abstract

Many patients with severe traumatic brain injuries (TBIs) undergo withdrawal of life-sustaining therapies (WLSTs) or transition to comfort measures, but noninjury factors that influence this decision have not been well characterized. We hypothesized that WLST would be associated with institutional and geographic noninjury factors. All patients with a head Abbreviated Injury Scale score ≥3 were identified from 2016 Trauma Quality Improvement Program data. We analyzed factors that might be associated with WLST, including procedure type, age, sex, race, insurance, Glasgow Coma Scale score, mechanism of injury, geographic region, and institutional size and teaching status. Adjusted logistic regression was performed to examine factors associated with WLST. Sixty-nine thousand fifty-three patients were identified: 66% male, 77% with isolated TBI, and 7.8% had WLST. The median age was 56 years (34-73). A positive correlation was found between increasing age and WLST. Women were less likely to undergo WLST than men (odds ratio 0.91 [0.84-0.98]) and took more time to for WLST (3 vs 2 days, P < .001). African Americans underwent WLST at a significantly lower rate (odds ratio 0.66 [0.58-0.75]). Variations were also discovered based on US region, hospital characteristics, and neurosurgical procedures. WLST in severe TBI is independently associated with noninjury factors such as sex, age, race, hospital characteristics, and geographic region. The effect of noninjury factors on these decisions is poorly understood; further study of WLST patterns can aid health care providers in decision making for patients with severe TBI.

ABBREVIATIONS:

AIS

Abbreviated Injury Scale

ICD

International Classification of Disease

NTDB

National Trauma Data Bank

TBIs

traumatic brain injury

TQIP

Trauma Quality Improvement Program

WLST

withdrawal of life-sustaining therapies.

Traumatic brain injury (TBI) represents approximately 16% of injury-related hospitalization and approximately one-third of trauma related mortality in the United States.¹ In severe cases of TBI, mortality rates have been reported as high as 35%, and severe TBI is associated with significant morbidity and functional limitations.^2,3 Studies estimate that withdrawal of life-sustaining therapies (WLST) occurs in more than 50% of severe TBI cases and that WLST is most often associated with severity of neurological injury.^4,5

Multiple studies demonstrate variability in end-of-life decision making. A qualitative study of 58 internal medicine physicians found that physician attitudes toward Do Not Resuscitate were reflective of institutional practices and attitudes. Trainees expressed the most discomfort with end-of-life decision making.⁶ Studies of standardized critically ill patients show that physician perceptions of WLST vary even at the provider level.⁷ Multiple studies have found variability in WLST rates across the world.^8-10

Few studies have investigated WLST trends in severe TBI. A study performed by Gambhir et al¹¹ identified patient-related and injury-related WLST factors, including age and race. However, noninjury factors for WLST in severe TBI have not been well characterized. The objectives of our study were to identify noninjury factors associated with WLST and to characterize differences in time to WLST for patients with severe TBI.

METHODS

We used the American College of Surgeons 2016 Trauma Quality Improvement Program (TQIP) 2016 National Trauma Data Bank (NTDB) to evaluate all patients with TBI admitted to over 850 participating trauma centers. We identified those patients who were reported to have suffered severe TBI as defined as a head Abbreviated Injury Scale (AIS) ≥3. From the initial 231 680 individuals recorded in the 2016 TQIP NTDB, this yielded 75 690 patients with severe TBI and 69 053 patients with complete WLST data for analysis (Figure). To evaluate those factors independently associated with WLST, we examined several patient-specific factors (age, sex, race, insurance type, Injury Severity Score, Glasgow Coma Scale [GCS], blunt or penetrating mechanism of trauma, and neurosurgical intervention type) and non–patient-specific factors (geographic region, hospital neurosurgeon group size, teaching status, trauma center level, and hospital size as determined by bed number). We recategorized race as White, Black, or African American and other categories for analysis purposes. Insurance status was consolidated to government, private, and other insurance categories. Government insurance refers to patients labeled as “Medicare,” “Medicaid,” or “other government insurance.”

FIGURE.

Methods. AIS, Abbreviated Injury Scale; ICD, International Classification of Disease; TBI, traumatic brain injury; TQIP, Trauma Quality Improvement Program.

Logistic regression was used to assess for those patient and nonpatient factors associated with WLST. Regressions were adjusted for age by decile, sex, race, insurance type, Injury Severity Score, GCS, blunt or penetrating mechanism of trauma, neurosurgical intervention type, geographic region, hospital neurosurgeon group size, teaching status, trauma center level, and hospital size as determined by bed number. To compare days to withdrawal, we first evaluated the normalcy of our data set using Shapiro–Wilk testing. This showed that the data were not distributed normally; thus, pairwise Wilcox rank sum testing was used for comparison of days with withdrawal of care across groups. Regression strength is reported as concordance statistic (C-stat = 0.868), regression results are reported as odds ratios and 95% CIs, and Wilcox rank sum testing is reported as median days with α = 0.05.

This study was approved by the Institutional Review Board. Patient consent was not required because the research presented no more than minimal harm to subjects and involved no procedures for which written consent is normally required. Statistics were performed using R, version 4.0.1 (R Foundation for Statistical Computing).

RESULTS

We included 69 053 patients in our study. Of those, 66% were male, 76.9% had an isolated TBI, and 3.8% had a penetrating mechanism of injury (Table 1). The median age was 59 years (36-77), and the median Injury Severity Score was 17 with 22.1% of patients having an emergency department GCS ≤8. Half of patients had a head AIS of 3 (52.7%) while less than 1% had a head AIS of 6. Our population consisted of 72.7% White and 10.6% Black or African American patients. Almost half of our patients had either government (52%) or private/commercial (42%) insurance. Patients were relatively evenly distributed around the United States, with majority in the South (37%). WLST was noted in 7.8% of patients (Table 1).

TABLE 1.

Demographics

	n = 69 053 (median [IQR] or % [n])
Age, y	59 (36-77)
Male sex	66.4% (n = 45 873)
Primary race identification
White	72.7% (n = 50 228)
African American	10.6% (n = 7304)
Asian	2.6% (n = 1818)
American Indian	0.9% (n = 594)
Native Hawaiian or Pacific Islander	0.3% (n = 183)
Others	8.3% (n = 5764)
Not specified	4.6% (n = 3162)
Hispanic ethnicity	10.9% (n = 7534)
Insurance status
Government	52.1% (n = 35 973)
Private/commercial	41.5% (n = 28 670)
Other/unknown	6.4% (n = 4410)
Injury Severity Score	17 (11-25)
Emergency department GCS ≤8	22.1% (n = 15 275)
Head AIS
3	51.7% (n = 35 724)
4	27.5% (n = 18 991)
5	20.5% (n = 14 187)
6	0.2% (n = 151)
Isolated TBI (all other body AIS <3)	76.9% (n = 53 121)
Penetrating mechanism	3.8% (n = 2632)
Region
Northeast	16.7% (n = 11 551)
Midwest	21.2% (n = 14 662)
West	22.6% (n = 15 612)
South	37.1% (n = 25 619)
Neurosurgeon group size
1-2	8.4% (n = 5827)
3-5	45.5% (n = 31 409)
≥6	46.1% (n = 31 817)

AIS, Abbreviated Injury Scale, GCS, Glasgow Coma Scale; TBI, traumatic brain injuries.

After adjusting for patient-related and injury-related factors, we saw a positive correlation between age and withdrawal of care (Table 2). As compared with individuals age younger than 19 years, an increase in odds of WLST was seen for every increasing decade of life. Individuals age 20 to 29 years were 1.26 times (odds ratio [OR] 1.01-1.59) more likely to have WLST while individuals age older than 90 years were 21.28 times (OR 16.55-27.50) more likely to have WLST. Regarding time to WLST, individuals age 30 to 79 years showed a significant increase in days to withdrawal of care (all P < .05), as compared with individuals younger than 19 years (Table 3). When examining sex, we found that women were less likely to undergo WLST than men (OR 0.91 [0.84-0.98]) (Table 2), as well as a significant increase in time to WLST (3 vs 2 days, P < .001) (Table 3). We also found that African Americans were 0.66 times (OR 0.58-0.75), almost half as likely, to withdraw care as their White counterparts. Similarly, other races were also less likely to withdraw care (OR 0.90 [0.81-1.00]) (Table 2). There was no significant difference in time to WLST regarding race (Table 3). Those with private insurance were less likely to undergo WLST (OR 0.86 [0.79-0.93]) (Table 2). There was no significant difference in time to WLST regarding insurance (Table 3). Patients with a GCS ≥9 were only 0.10 times as likely to undergo WLST (OR 0.09-0.11) than those with a GCS of <9 (Table 2) and demonstrated a significant increase to time to WLST (5 vs 2, P < .001) (Table 3). Finally, patients with a penetrating mechanism of injury were 2.59 times (OR 2.26-2.95) more likely to undergo WLST (Table 2) and underwent WLST faster (1 vs 3 days, P < .001) (Table 3).

TABLE 2.

Regression Analysis of Injury and Noninjury Factors on Odds to WLST

Reference group	Factor	Odds for WLST
		OR	2.50%	97.50%
Age 16-19, y	20-29	1.26	1.01	1.59
	30-39	1.60	1.27	2.03
	40-49	2.35	1.86	2.99
	50-59	4.34	3.49	5.44
	60-69	6.72	5.39	8.43
	70-79	13.51	10.81	17.01
	80-89	20.77	16.59	26.20
	≥90	21.28	16.55	27.50
Male	Female	0.91	0.84	0.98
Caucasian	African American	0.66	0.58	0.75
	Others	0.90	0.81	1.00
Government insurance	Other/unknown	0.97	0.82	1.13
	Private	0.86	0.79	0.93
GCS <9	GCS ≥9	0.10	0.09	0.11
Nonpenetrating	Penetrating	2.59	2.26	2.95
Midwest region	North East	1.04	0.93	1.16
	South	0.87	0.79	0.95
	West	0.85	0.76	0.94
Neurosurgeon small (1-2) group	Large group (≥6)	1.08	0.95	1.24
	Medium group (3-5)	0.99	0.87	1.13
Teaching status: University	Community	1.08	0.99	1.18
	Nonteaching	1.13	0.97	1.30
Trauma center level 1	Trauma level 2	0.87	0.78	0.96
	Trauma level 3	1.32	0.72	2.29
Hospital size <200 beds	201-400 beds	0.85	0.71	1.02
	401-600 beds	0.71	0.59	0.87
	>600 beds	0.71	0.59	0.86
No intervention	Craniotomy and ICP	2.03	1.68	2.44
	Craniotomy only	1.41	1.21	1.63
	ICP only	2.15	1.93	2.40

GCS, Glasgow Coma Scale, ICP, intracranial pressure; OR, odds ratio; WLST; withdrawal of life-sustaining therapies.

TABLE 3.

Injury and Noninjury Factors Effect on Time to WLST

Reference group	Factor	Days to WLST
		Median daysa	P-value
Age 16-19, y	20-29	2(4) vs 2(5)	.28
	30-39	2(4) vs 3(6)	.04
	40-49	2(4) vs 3(8)	<.001
	50-59	2(4) vs 4(8)	<.001
	60-69	2(4) vs 3(8)	<.001
	70-79	2(4) vs 3(7)	.01
	80-89	2(4) vs 2(6)	.12
	≥90	2(4) vs 2(4)	.93
Male	Female	3(7) vs 2(5)	<.001
Caucasian	African American	3(6) vs 3(7.5)	.23
	Others	3(6) vs 3(7)	.23
Government insurance	Other/unknown	3(6) vs 3(6)	.8
	Private	3(6) vs 3(7)	.31
GCS < 9	GCS ≥9	2(5) vs 5(8)	<.001
Nonpenetrating	Penetrating	3(7) vs 1(2)	<.001
Midwest region	North East	2(5) vs 3(7)	.17
	South	2(5) vs 3(6)	<.001
	West	2(5) vs 3(6)	<.001
Neurosurgeon small (1-2) group	Large group (≥ 6)	3(6) vs 3(6)	.94
	Medium group (3-5)	3(6) vs 3(6)	.89
Teaching status: University	Community	3(7) vs 3(6)	.4
	Nonteaching	3(7) vs 3(6)	.38
Trauma center level 1	Trauma level 2	3(6) vs 3(6)	.27
	Trauma level 3	3(6) vs 1(2)	.08
Hospital size <200 beds	201-400 beds	3(6) vs 3(6)	.94
	401-600 beds	3(6) vs 3(6)	.83
	>600 beds	3(6) vs 3(7)	.46
No intervention	Craniotomy and ICP	2(5) vs 7(7.5)	<.001
	Craniotomy only	2(5) vs 5(8)	<.001
	ICP only	2(5) vs 6(7)	<.001

GCS, Glasgow Coma Scale; ICP, intracranial pressure; WLST; withdrawal of life-sustaining therapies.

^aValues are reported as median (IQR).

After adjusting for injury-related factors, we found several non–injury-related factors associated with WLST. The South and West were less likely to withdraw care than the Midwest (OR 0.87 [0.79-0.95] and OR 0.85 [0.76-0.94], respectively) (Table 2). Time to WLST was longer in both the South and West (3 vs 2 days, P < .001 for both) (Table 3). No significant difference was found in time to WLST based on neurosurgeon group size (P > .05) (Table 3). There was no significant difference in odds for WLST or time to WLST based on hospital teaching status. When compared with level 1 trauma center status, level 2 centers were less likely to WLST (OR 0.87 [0.78-0.96]) while level 3 centers showed no significant difference (OR 1.32 [0.72-2.29]) (Table 2). There was no significant difference in time to WLST based on trauma center level (Table 3). Finally, as compared with a hospital size of <200 beds, hospitals with 401 to 600 and >600 beds were both less likely to WLST (OR 0.71 [0.59-0.87] and OR 0.71 [0.59-0.86], respectively) (Table 2). No significant difference was found in time to WLST based on hospital bed size (Table 3). Patients who had craniotomy, intracranial pressure monitor placement, or both were all more likely to undergo WLST (OR 1.41 [1.21-1.63], OR 2.15 [1.93-2.40], and OR 2.03 [1.68-2.44], respectively) (Table 2). Time to WLST was significantly longer in all groups (Table 3). The median time to WLST for those who underwent craniotomy only was 5 days (P < .001), intracranial pressure only 6 days (P < .001), and both procedures 7 days (P < .001), compared with 2 days in the no intervention group (Table 3).

DISCUSSION

Key Results

Our results demonstrate that both injury and noninjury patient factors were associated with WLSTs in patients with severe TBI. The association of age with WLST was expected because TBI outcomes are known to be worse with age.¹² An increase in time to WLST was found in age groups ranging from 30 to 79 years relative to patients younger than 19 years. This finding likely reflects the age-specific, bimodal distribution of injury severity and mortality in TBI, which shows the highest incidence and severity in late adolescent/early adulthood and again in the elderly population.² When examining sex, we found that women were less likely to undergo WLST, and, when they did, the time to withdrawal was significantly longer than that for men. This was similar to a 2013 multicenter cohort study that had an approximately 70% male-dominated population for patients who had WLSTs in patients with severe TBI.¹³

Interpretation

Our findings build on the literature involving patient demographic, injury characteristics, and WLST. African Americans were almost half as likely to withdraw care. There was a similar finding in a 2001 single-institution study investigating factors for withdrawal of care, which found that mechanical ventilation was less likely to be withdrawn in patients who were African American.¹⁴ This may be reflective of a lack of trust in the medical community and cultural or religious beliefs that require further investigation to identify differences across race.^15-18 Those with a higher GCS were significantly less likely to undergo WLST, which was expected given the previous literature on GCS and prognosis.¹⁹ Finally, we found that insurance status did not seem to affect WLST decisions, which is consistent with previous findings in neurological injury.⁴

Our study showed variation in care based on region of the United States and hospital factors, which has been demonstrated in other disease processes and procedures.^20-23 The Midwest and Northeast regions showed the highest odds for WLST, supporting that regional culture may be influencing these decisions. These findings are consistent with previous studies in WLST, which showed regional, institutional, and even interprovider variation in WLST rates.^8,24 Further investigation is warranted to understand regional bias in decision-making patterns for WLST.²⁵ Larger hospitals and level 2 trauma centers also showed significant differences in WLST, which may also be due to local and institutional culture and practices. Finally, we saw an increase in withdrawal of care after procedures, which is most likely associated with the severity of TBI.

Many studies have proposed changes to policy and institutional guidelines as a way to improve end-of-life decision making.^26-28 Medical students and residents have expressed low comfort and little training regarding end-of-life conversations, pointing to a foundational problem in medical education.²⁹ Research shows that many end-of life conversations in a hospital setting take place without an attending and that only a quarter of trainees were comfortable having these conversations.³⁰ Moreover, the process of WLST decision making in TBI is distinctly challenging with decisions made under time-limited circumstances often without a clear understanding of patient preferences regarding life-long neurological and functional deficits.^5,31 These decisions are often influenced by cognitive biases in the acute care period that have been previously described and are known to affect treatment decisions and outcomes in the setting of acute neurological injury.³² Continued study of noninjury factors and biases that affect decision making in TBI can help to educate trainees and improve decision making for patients with TBI.

Limitations

This study uses a retrospective data set, and, as such, our findings can only identify associations without establishing causation. The large number of trauma centers that contributed to the 2016 TQIP NTDB makes our findings broadly generalizable. We combined Medicare and Medicaid insurance status into a single referent group, which limits any detailed conclusions about the association of insurance status and WLST. However, previous studies suggest that insurance status is not strongly associated with WLST decisions for patients with neurological injury.⁴

This study makes use of AIS designation to define TBI severity as opposed to relying solely on GCS criteria. Multiple studies have shown the problems associated with the use of GCS alone as a marker for prognosis and TBI severity including its unreliability.^33-35 Head AIS scoring is commonly used to define injury severity when using trauma registry data and to benchmark trauma quality and outcomes. Head AIS score ≥3 has been used over time in the trauma literature to define severe TBI.^36,37 We have previously shown that head AIS scores ≥3 can define a population of severe TBI that would not be predicted by GCS alone.^38,39 Most often, the use of both GCS and head AIS can give a more complete picture of TBI severity. There remains some debate about how to best define severe TBI when using trauma registry data.^35,40 In our study, head AIS criteria allowed us to identify all patients with TBI who might be associated with a withdrawal of care episode.

CONCLUSION

WLST in severe TBI is independently associated with noninjury factors such as sex, age, race, hospital characteristics, and geographic region. The effect of noninjury factors on these decisions is poorly understood; further study of WLST patterns can aid health care providers in decision making for patients with severe TBI.

Funding

This publication was made possible by the Clinical and Translational Science Collaborative of Cleveland, KL2TR002547 (to Dr Ho) from the National Center for Advancing Translational Sciences (NCATS) component of the National Institutes of Health and NIH roadmap for Medical Research.

Disclosures

The authors have no personal, financial, or institutional interest in any of the drugs, materials, or devices described in this article.