Differences in code status and end of life decision making in patients with limited English proficiency in the intensive care unit

Abstract

Objective:

The purpose of this study was to determine if code status, advance directives, decisions to limit life support were different for ICU patients with LEP compared with patients whose primary language was English.

Methods:

Retrospective cohort study in 7 adult ICUs in a single tertiary academic medical center May 31 2011-through June 1 2014.

Results:

Out of 27,523 patients admitted to the ICU, 779 (2.8%) had LEP. When adjusted for severity of illness, sex, education, and insurance status, patients with LEP were less likely to change their code status from full code to do not resuscitate (DNR) during ICU admission (OR, 0.62; 95% CI, 0.46–0.82; p<0.001) and took 3.8 days (95% CI, 1.9–5.6; p<0.001) longer to change to DNR. Patients with LEP who died in the ICU were less likely to receive a comfort measures order set (OR, 0.38; 95% CI 0.16–0.91; p=0.03) and took 19.1 days (95% CI, 13.2–25.1; p<0.001) longer to transition to comfort measures only. LEP patients were less likely to have an advance directive (OR, 0.23; 95% CI, 0.18–0.29; p<0.001), more likely to receive mechanical ventilation (OR, 1.26; 95% CI, 1.07–1.48; p=0.005), and more likely to have restraints used (OR, 1.36; 95% CI 1.11–1.65; p=0.003). Hospital length of stay was 2.7 days longer for LEP patients. Additional adjustment for religion, race and age yielded similar results.

Conclusion:

There are important differences in end-of-life care and decision-making for patients with LEP.

Introduction

Approximately one out of every 12 adults in the United States has limited English proficiency (LEP), according to 2013 United States Census Bureau estimates.¹ Between 1990 and 2013, the adult population with LEP grew from 6.1% (14.0 million) to 8.5% (25.1 million). While a majority of the LEP population in the US speaks Spanish (64%), up to 350 languages are spoken.¹ Limited English proficiency is defined as “not speaking English as a primary language and potentially having a limited ability to read, speak, write or understand English.”². Language barriers are associated with numerous adverse health outcomes such as longer hospital stays,³ increased rates of hospital readmission,⁴ lower rates of understanding discharge instructions,⁵ lower satisfaction,^6–8 lower rates of receiving recommended preventative medical services,^{9, 10} decreased rates of medication adherence,¹¹ deferring needed medical care,¹² lower rates of having a primary care provider,¹³ compromised patient-physician communication,¹⁴ and higher healthcare utilization and cost.¹⁵

Because communication is an essential component of decision making at the end of life, patients with LEP are at specific risk for suboptimal decision making. In the outpatient setting, patients with language barriers are at risk for lower quality of end of life care, suboptimal discussions about goals of care, not having an accurate understanding of their diagnoses and prognoses, and receiving suboptimal symptom control.^16–18 Family members of intensive care unit (ICU) patients with LEP are at risk for receiving less information and less emotional support during ICU family conferences.¹⁹ Knowledge gaps exist regarding the impact of LEP on decision making for life support and at the end of life for hospitalized patients in the ICU. Understanding such knowledge gaps is critical to identifying methods to improve communication and decision making.²⁰ The objective of this study was to determine if code status, advance directives, life support preferences, use of comfort measures prior to death, and timing of decision making were different for ICU patients with LEP compared to patients who spoke English.

Methods

Setting and study design:

We conducted a retrospective cohort study of all adult patients admitted to seven ICUs (medical, neurologic, cardiac, mixed, and 3 surgical) in a single center tertiary academic medical center from May 31, 2011 to June 1, 2014. The Mayo Clinic Institutional Board of Review approved the study protocol. Inclusion characteristics were patients ≥ 18 years old who were admitted to the ICU and gave research authorization. There was no contact with patients.

Limited English proficiency definition:

Limited English proficiency was defined as a primary language other than English². This variable was abstracted from the electronic medical record using an automated retrieval query. The accuracy of the automated retrieval query was verified by manual abstraction of a sample of 100 records. LEP status for patients whose primary language or interpreter use was unknown was verified by manual chart abstraction.

Data collection:

The following demographic variables were abstracted for each patient from registration data in the electronic medical record: age, sex, race/ethnicity, marital status, religion, education level and insurance type. Medical complexity was assessed by the Charlson co-morbidity index, which considers the number and severity of 19 predefined comorbid conditions (as identified by ICD-9 codes) and provides a weighted score of a patient’s comorbidities.²¹ The Acute Physiology and Chronic Health Evaluation (APACHE) III score was calculated for each patient to assess illness severity upon admission to the ICU and 24 hours after admission.²²

The primary outcomes of the study were characteristics of decision-making for life support, code status, and aggressiveness of treatment²³ and included: code status on ICU admission, code status on ICU discharge, change in code status during ICU stay, use of life support (invasive mechanical ventilation, noninvasive mechanical ventilation, dialysis, vasopressors, CPR), presence of advance directives, and implementation of a standardized institutional comfort measures only order set. Secondary outcomes included use of restraints, documentation of a family conference, the presence of symptoms²³ (delirium, pain, and agitation), ICU and hospital length of stay and mortality, and hospital discharge location. All outcomes were collected from the electronic medical record using automated retrieval queries. For patients for whom the presence of advance directives was not available using an automated search query, manual chart abstraction and imputation strategies were utilized.

Presence of pain was defined as either a pain rating of 3 or greater on a 0-to-10 numeric pain intensity scale or a FLACC (Face, Legs, Activity, Cry, and Consolability) score of 5 or greater in the final 24 hours of ICU stay.²⁴ Agitation was defined as RASS (Richmond Agitation-Sedation Scale) of +2 or greater during the final 24 hours of ICU stay.²⁵ Confusion was defined as CAM-ICU (Confusion Assessment Method for the ICU) positive in the final 24 hours of ICU stay.²⁶

Statistical analysis:

All continuous variables were reported as medians with interquartile range (IQR). In order to minimize the effects of outliers and variables with non-normal distributions, Wilcoxon rank-sum tests were used to compare continuous variables between groups. We performed multivariate linear regression for continuous outcomes including time to code status change, time on mechanical and non-invasive mechanical ventilation, hospital and ICU lengths of stay, and time to comfort measures only order, adjusting for APACHE III score, sex, education, and insurance. Parameter estimates with 95% confidence intervals were reported. All categorical variables were reported as counts with percentages and compared between groups using chi-square tests. Binary outcomes were also analyzed using multivariate logistic regression to adjust for APACHE III score, sex, education, and insurance. The adjusted odds ratios with 95% confidence intervals were reported. For patients for whom the presence of advance directives was not available using an automated EMR search query, manual chart abstraction and imputation strategies were utilized. A two-sided p value of less than.05 was considered statistically significant. All analyses were done using SAS version 9 in a Linux environment.

Sensitivity Analysis was performed on all outcomes adjusting for additional variables including race, religion and age (as well as 24 hour APACHE III score, education, insurance but not sex).

Results

Baseline Characteristics:

A total of 27,523 patients were admitted to the ICU from May 31, 2011 to June 1, 2014. Of these, 779 (2.8%) had limited English proficiency and 26,744 (97.2%) did not have limited English proficiency (Table 1). Patients with LEP spoke the following languages: Arabic (26.4%), Spanish (26.3%), Somali (8.7%), Cambodian (4.4%), Vietnamese (2.8%), Lao (2.6%), Hmong (2.4%), American Sign Language (2.3%), Russian (2.1%), and other languages (22.0%).

Table 1.

Baseline Characteristics of patient population

Characteristic	Limited English proficiency (n=779)	No Limited English proficiency (n=26744)	p value
Age, years, n (%)			<.001
<40	204 (26.2%)	3181 (11.9%)
40–49	92 (11.8%)	2654 (9.9%)
50–59	147 (18.9%)	4758 (17.8%)
60–69	165 (21.2%)	6143 (23.0%)
70–79	108 (13.9%)	5644 (21.1%)
≥ 80	63 (8.1%)	4364 (16.3%)
Female sex, n (%)	340 (43.6%)	11545 (43.2%)	.79
Race, n (%)			<.001
White	266 (34.1%)	25249 (94.4%)
Black or African American	78 (10.0%)	388 (1.5%)
Asian	120 (15.4%)	206 (0.8%)
American Indian or Alaska Native	2 (0.3%)	139 (0.5%)
Other	236 (30.3%)	408 (1.5%)
Unknown	77 (9.9%)	354 (1.3%)
Ethnicity, n (%)			<.001
Hispanic	176 (22.6%)	317 (1.2%)
Non-Hispanic	514 (66.0%)	25173 (94.1%)
Unknown	89 (11.4%)	1254 (4.7%)
Insurance, n (%)			<.001
Medicaid Only	173 (22.2%)	2061 (7.7%)
Medicare Only	82 (10.5%)	7573 (28.3%)
Medicare Plus Private	44 (5.6%)	5132 (19.2%)
Private Insurance Only	366 (47.0%)	11409 (42.7%)
Uninsured	114 (14.6%)	569 (2.1%)
Marital status, n (%)			<.001
Single	277 (35.6%)	9810 (36.7%)
Married or long term partner	464 (59.6%)	16726 (62.5%)
Unknown	38 (4.9%)	208 (0.8%)
Religion, n (%)			<.001
Buddhist	38 (4.9%)	34 (0.1%)
Christian	298 (38.3%)	20685 (77.3%)
Muslim	271 (34.8%)	78 (0.3%)
Jewish	18 (2.3%)	214 (0.8%)
Hindu	6 (0.8%)	38 (0.1%)
Other	83 (10.7%)	3785 (14.2%)
Unknown	65 (8.3%)	1910 (7.1%)
Education, n (%)
Some High School or Less	241 (30.9%)	2166 (8.1%)	<.001
High School Graduate	147 (18.9%)	8210 (30.7%)
Any College	70 (9.0%)	6634 (24.8%)
College Graduate	140 (18.0%)	6479 (24.2%)
Unknown	181 (23.2%)	3255 (12.2%)
Charlson comorbidity index, median (IQR)	1.0 (0–3)	1.0 (0–3)	.90
APACHE III score, median (IQR) 1 hour	36 (24–51)	40 (28–55)	<.001
APACHE III score, median (IQR) 24 hours	54(41–73)	58(42–75)	<.001
Primary language spoken			<.001
English	0 (0%)	26744 (100%)
Arabic	206 (26.4%)	0 (0%)
Spanish	205 (26.3%)	0 (0%)
Somali	68 (8.7%)	0 (0%)
Cambodian (Khmer)	34 (4.4%)	0 (0%)
Vietnamese	22 (2.8%)	0 (0%)
Lao	20 (2.6%)	0 (0%)
Hmong	19 (2.4%)	0 (0%)
American Sign Language	18 (2.3%)	0 (0%)
Russian	16 (2.1%)	0 (0%)
Other	171 (22.0%)	0 (0%)

^aAPACHE=Acute Physiology and Chronic Health Evaluation; IQR=interquartile range

Patients with LEP were more likely to be younger (median 56.6 versus 64.7 years, p<.001), be uninsured (14 % versus 2.1%, p<.001), and have a lower APACHE III score (54 versus 58, p<.001). LEP patients were less likely to have a White race (34.1% versus 94.4%, p<.001), report Christian religion (38.3% versus 77.3%, p<.001), or be high school graduates (18.9% versus 30.7%, p<.001). The sex and Charlson co-morbidity scores were not significantly different between the two groups.

Code status:

Patients with LEP were less likely to have a DNR order on ICU discharge (7.3% versus 10.3%, p=.007), and were less likely to change from full code to DNR during ICU admission (6.8% versus 9.4%, p=.02). Patients with LEP were statistically less likely to have a do not resuscitate (DNR) order on ICU admission (0.5% versus 1.5%, p=.02)., Among patients who did have a change in code status from full code to DNR, the median number of days to make the change was 15.7 for LEP patients compared to 3.3 for non LEP patients, p=.01 (Table 2). When adjusted for APACHE III score, sex, education level, and insurance status, the odds ratio (95% CI) for DNR on ICU admission for LEP patients was 0.30 (0.11–0.80), p=.02 and for DNR on ICU discharge for LEP patients was 0.60 (0.45–0.79), p<.001 (Table 3). The adjusted parameter estimate (95% CI) for days from ICU admission to code status change to DNR was 3.8 days (1.9–5.6) longer for LEP patients than non-LEP patients (p<.001) (Table 4).

Table 2.

Unadjusted outcomes in patients with limited English proficiency compared to no limited English proficiency

Outcome	Limited English proficiency (n=779)	No limited English proficiency (n=26744)	p value
Code status and advance directives
DNR on ICU admission, n (%)	4 (0.5%)	403 (1.5%)	.02
DNR on ICU discharge, n (%)	57 (7.3%)	2750 (10.3%)	.007
Change from Full Code to DNR, n (%)	53 (6.8%)	2506 (9.4%)	.02
Days from ICU admission to change to DNR, median (IQR)	15.7 (1.9–64.6)	3.3 (1.0–25.9)	.01
Advance directives present at ICU admission, n (%)	86 (11.0%)	9842 (36.8%)	<.001
Documentation of a family conference, n (%)	49 (6.3%)	672 (2.5%)	<.001
Life support and other treatment utilization
Mechanical ventilation use, n (%)	292 (37.5%)	9642 (36.1%)	.41
Mechanical ventilation days, median (IQR)	0.4 (0.2–1.5)	0.4 (0.2–1.2)	.15
Noninvasive ventilation use, n (%)	72 (9.2%)	3532 (13.2%)	.001
Noninvasive ventilation days, median (IQR)	0.5 (0.1–1.2)	0.5 (0.2–1.4)	.47
Dialysis use, n (%)	54 (6.9%)	1538 (5.8%)	.16
Vasopressor use, n (%)	188 (24.1%)	6453 (24.1%)	.10
CPR performed, n (%)	3 (0.4%)	94 (0.4%)	.88
Continuous IV analgesia used, n (%)	83 (10.7%)	3122 (11.7%)	.38
Continuous IV sedation used, n (%)	112 (14.4%)	4821 (18.0%)	.009
Restraints used, n (%)	125 (16.0%)	3202 (12.0%)	<.001
Patient Symptoms
Delirium, n (%)	32 (4.1%)	1790 (6.7%)	.004
Pain, n (%)	439 (56.4%)	17589 (65.8%)	<.001
Agitation, n (%)	28 (3.6%)	1010 (3.8%)	.79
Length of stay and disposition
ICU length of stay, days, median (IQR)	1.2 (0.9–2.6)	1.1 (0.9–2.3)	.76
Hospital length of stay, days, median (IQR)	6.2 (3.3–10.5)	5.4 (3.3–9.0)	.002
Hospital discharge to location other than home, n (%)	100 (13.2%)	6141 (23.7%)	<.001
ICU mortality	22 (2.8%)	771 (2.9%)	.92
Hospital mortality	36 (4.6%)	1300 (4.9%)	.76
Deaths in the ICU	n=22	n=771
Comfort measures only order placed, n (%)	9 (40.9%)	499 (64.7%)	.02
Days from ICU admission to comfort care order, median (IQR)	2.4 (1.5–16.0)	2.5 (0.9–6.4)	.30
Chaplain visitation, n (%)	0 (0%)	21 (2.7 %)	.43
Palliative Care consultation, n (%)	6(27.3%)	176(22.8%)	.62
Delirium within 24 hours of death, n (%)	2 (9.1%)	105 (13.6%)	.54
Pain within 24 hours of death, n (%)	1 (4.5%)	106 (13.7%)	.21
Agitation within 24 hours of death, n (%)	1 (4.5%)	60 (7.8%)	.57

^aCPR=cardiopulmonary resuscitation DNR=do not resuscitate; ICU=intensive care unit; IQR=interquartile range; IV=intravenous;

Table 3:

Adjusted outcomes in patients with limited English proficiency compared to no limited English proficiency (categorical variables)

Outcome	Odds ratio, unadjusted (95% CI)	p value	Odds ratio, adjusted* (95% CI)	p value
Code status and advance directives
DNR on ICU admission	0.34 (0.13,0.91)	.03	0.30 (0.11,0.80)	.02
DNR on ICU discharge	0.69 (0.53,0.91)	.007	0.60 (0.45,0.79)	<.001
Change from Full Code to DNR	0.71 (0.53,0.94)	.02	0.62 (0.46,0.82)	<.001
Advance directives present at ICU admission	0.21 (0.17,0.27)	<.001	0.23 (0.18,0.29)	<.001
Documentation of a family conference	2.60 (1.93,3.51)	<.001	2.53 (1.86,3.43)	<.001
Life support and other treatment utilization
Mechanical ventilation use	1.06 (0.92,1.23)	.41	1.26 (1.07,1.48)	.005
Noninvasive ventilation use	0.67 (0.52,0.86)	.001	0.66 (0.52,0.85)	.001
Dialysis use	1.22 (0.92,1.62)	.16	1.29 (0.97,1.71)	.09
Vasopressor use	1.00 (0.85,1.18)	.99	1.14 (0.95,1.36)	.16
CPR performed	1.10 (0.35,3.47)	.88	1.07 (0.34,3.40)	.91
Continuous IV analgesia used	0.90 (0.72,1.14)	.38	1.03 (0.82,1.31)	.78
Continuous IV sedation used	0.76 (0.62,0.94)	.009	0.88 (0.71,1.08)	.21
Restraints used	1.41 (1.16,1.71)	<.001	1.36 (1.11,1.65)	.003
Patient Symptoms
Delirium	0.60 (0.42,0.85)	.005	0.56 (0.39,0.81)	.002
Pain	0.67 (0.58,0.78)	<.001	0.73 (0.63,0.84)	<.001
Agitation	0.95 (0.65,1.39)	.80	0.87 (0.59,1.28)	.47
Length of stay and disposition
Hospital discharge to location other than home	0.61 (0.52,0.73)	<.001	0.57 (0.49,0.68)	<.001
ICU mortality	0.98 (0.64,1.51)	.92	0.90 (0.58,1.40)	.65
Hospital mortality	0.95 (0.68,1.33)	.76	0.88 (0.62,1.25)	.48
Deaths in the ICU
Comfort measures only order placed	0.38 (0.16,0.89)	.03	0.38 (0.16,0.91)	.03
Palliative Care consultation	1.27 (0.49,3.29)	.63	1.29 (0.49,3.40)	.61
Delirium within 24 hours of death	0.63 (0.15,2.75)	.54	0.65 (0.15,2.84)	.57
Pain within 24 hours of death	0.30 (0.04,2.24)	.24	0.27 (0.04,2.04)	.20
Agitation within 24 hours of death	0.56 (0.07,4.27)	.58	0.55 (0.07,4.21)	.57

^aCPR=cardiopulmonary resuscitation DNR=do not resuscitate; ICU=intensive care unit; IQR=interquartile range; IV=intravenous;

^*adjusted for APACHE III score, sex, education, and insurance

Table 4:

Adjusted outcomes in patients with limited English proficiency compared to no limited English proficiency (continuous variables)

Outcome	Parameter estimate, unadjusted (95%CI)	p value	Parameter estimate, adjusted* (95% CI)	p value
Days from ICU admission to change to DNR	3.6 (1.7,5.4)	<.001	3.8 (1.9,5.6)	<.001
Mechanical ventilation days	0.3 (0.1, 0.5)	.002	0.3 (0.1, 0.5)	<.001
Noninvasive ventilation days	−0.1 (−0.1,−0.0)	.01	−0.1 (−0.1,−0.0)	.01
ICU length of stay, days	0.6 (0.3, 1.0)	<.001	0.6 (0.3, 1.0)	<.001
Hospital length of stay, days	2.5 (1.8, 3.3)	<.001	2.7 (2.0, 3.5)	<.001
Days from ICU admission to comfort measures only order	19.3 (13.3–25.4)	<.001	19.1 (13.2,25.1)	<.001

^*adjusted for APACHE III score, sex, education, and insurance

^aCI=confidence interval; DNR= do not resuscitate ICU=intensive care unit;

Advance directives:

Patients with LEP were less likely to have advance directives present during ICU admission (11.0% versus 36.8%, p<.001) (Table 2). The adjusted odds ratio (95% CI) for advance directives for LEP patients was 0.23 (0.18–0.29), p<.001 (Table 3). Documentation of a family conference was higher for LEP patients with an adjusted odds ratio (95% CI) of 2.53 (1.86–3.43), p<.001 (Table 3).

Life support and other treatment utilization:

There were no differences in the unadjusted rates of mechanical ventilation between the LEP and non LEP groups (37.5% and 36.1%), (Table 2). But after adjusting for APACHE III score, sex, education level, and insurance status, the odds ratio (95% CI) for using mechanical ventilation for LEP patients was 1.26 (1.07–1.48), p=.005 (Table 3). The adjusted parameter estimate for length of mechanical ventilation was 0.3 days longer for LEP patients (Table 4)., Noninvasive ventilation was used less in LEP patients (9.2% versus 13.2%), p=.001 (Table 2) with an adjusted odds ratio (95% CI) of 0.66 (0.52–0.85), p=.001 (Table 3). When using adjusted estimates, there were no differences in rates of dialysis, vasopressors, CPR, continuous intravenous analgesia, or continuous intravenous sedation. Restraints were used more often in patients with LEP (16.0% versus 12.0%) with an adjusted odds ratio (95% CI) of 1.36 (1.11–1.65), p=.003.

Nursing reported patient symptoms

Rates of nurse-assessed delirium and pain were lower for LEP patients (56.4% versus 65.8%, and 4.1% versus 6.7%, respectively). These differences persisted after adjustment. Rates of agitation did not differ (3.6% versus 3.8%).

Length of stay, mortality, and disposition

The unadjusted median ICU length of stay for LEP patients and English speaking patients was 1.2 days and 1.1 days respectively. The unadjusted median hospital length of stay was 6.2 days for LEP patients and 5.4 days for English speaking patients, p=.002 (Table 2). The adjusted ICU length of stay (95% CI) was 0.6 days (0.3–1.0) longer for LEP patients (p<.001) and the adjusted hospital length of stay (95% CI) was 2.5 days (1.8–3.3) longer for LEP patients (p<.001). No differences were noted in the unadjusted or adjusted ICU or hospital mortality rates. Patients with LEP were more likely to be discharged home following ICU admission compared to English speaking patients. The OR (95% CI) for discharge to location other than home, (such as a skilled nursing facility) for LEP patients was 0.57 (0.49–0.68), p<.001.

Comfort measures prior to death

Among the 793 patients who died in the ICU, 22 (2.8%) had LEP and 771 (97.2%) spoke English. An order for comfort measures only was utilized before death in 40.9% of patients with LEP versus 64.7% of patients who spoke English, p=.02 (Table 2). The adjusted OR (95% CI) for use of comfort measures only prior to death was 0.38 (0.16–0.91), p=.03 for LEP (Table 3). Among patients for whom a comfort measures only was utilized, it took 19.1 days longer to place the order for LEP patients than for English speaking patients, p<.001 (Table 4). There were no differences in the rates of palliative care consultation or recorded pain, delirium, or agitation in the 24 hours prior to death.

Sensitivity Analysis

During the sensitivity analysis we found that some differences documented above no longer reached statistical significance,(supplemental Table) suggesting that some of the differences may have been influenced by religion and race as much as limited English proficiency. However LEP remained independently associated with lower use of advance directives and lower rates of do-not-resuscitate orders in elderly patients. LEP patients who switched to do-not-resuscitate in the ICU, took significantly longer to switch. Patients with LEP had longer ICU stays and hospital stays, and received more restraints.

Discussion

To our knowledge, this is the first study to show that patients with LEP in the ICU had lower rates of do-not-resuscitate orders, lower rates of comfort measure orders use before death, and lower rates of advance directive completion on ICU admission. When do-not-resuscitate and comfort measures orders before death were utilized, it took significantly longer to reach these decisions for patients with LEP than for English speaking patients. LEP patients had longer hospital stays although mortality outcomes in the ICU and hospital were comparable. Use of restraints was noted to be statistically higher among those with LEP despite no differences in rates of agitation, with an adjusted OR (95% CI) of 1.36 (1.11–1.65, p=.003).

Patients who did not speak English had lower rates of available completed advance directives on ICU admission compared to patients who spoke English—and this likely represents a true disparity in medical care. Although advance care planning has many limitations, it is one cornerstone of end of life decision making in the United States.^{27, 28} Patients who completed advance directives have been shown to be more likely to receive medical care that was in agreement with their stated preferences than patients without advance directives.^{29, 30} Thus, it is possible that lower rates of advance directive completion for patients with language barriers potentially represents the delivery of medical treatment that does not fully honor their preferences. While numerous studies have shown lower rates of advance directive completion for minority groups based on race, ethnicity, and culture, this is the first to show lower completion rates based on limited English proficiency.^{31, 32} Qualitative studies have suggested that communication barriers may contribute to lack of effective advance care planning among vulnerable populations.^{33, 34}

Patients with LEP were less likely to have documentation of a comfort measures only order set and if the order set was utilized, it took significantly longer to implement the order set compared to patients who spoke English well. This finding persists despite similar rates of palliative care consultation and increased rates of documentation of family conferences for patients with LEP. Utilization of a comfort measures only order set prior to death explicitly reflects both a decision to withhold or withdraw unnecessary medical treatments aimed to extend life (such as life support, cardiopulmonary resuscitation, vital sign monitoring, etc.) as well as a deliberate decision to implement therapies aimed to relieve symptoms common at the end of life (such as pain, dyspnea, nausea, anxiety, etc.).³⁵ Implementing comfort measures only prior to death is a recommended component of providing optimal palliative care at the end of life in hospitalized patients.^{36, 37} While studies have shown that African American and Hispanic populations are more likely to choose life support, this study is among the first to document life support preferences and code status for hospitalized patients with LEP.³¹

Lower rates of and delayed initiation of comfort measures prior to death and lower rates of DNR orders in patients with LEP may reflect two possibilities: 1) more patients with LEP have an authentic desire to die with full medical therapies rather than a comfort measures only approach or 2) communication or other barriers prevent health care teams from optimally assessing and implementing a comfort measures approach for dying patients with LEP. The vast cultural, linguistic, and religious heterogeneity of this sample implies that communication barriers between healthcare teams and patients may be a primary driver of these differences. This is consistent with previous studies documenting numerous barriers to optimal communication and decision making between health care teams and patients with LEP.¹⁴ Video recordings and surveys of patient/physician interactions shows that patients with LEP receive less information, speak less during clinical encounters, are more likely to have their comments ignored, and are less satisfied with communication.^{14, 38} Family members of patients who use medical interpreters receive less information and emotional support during ICU family conferences.¹⁹

Physician lack of time and increased workload are additional factors that may contribute to suboptimal communication and end of life decision making in the ICU.³⁹ Furthermore, communication with a patient with LEP or a patient who requires an interpreter may be a barrier for ICU physicians. Results of previous studies suggest that physicians may not see the benefit of professional interpreters, may get frustrated with interpreters, and may put the blame for needing an interpreter on the patient.⁴⁰ Further studies are needed to clarify goals of care discussions among patients with limited English proficiency in the ICU. When sensitivity analysis was done on the cohort by adjusting for race, religion and age (as well as APACHE III score, education, insurance but not sex) patients with LEP were still noted to have lower rates of advance directives, took longer to reach DNR and have longer hospital and ICU lengths of stay. Elderly LEP patients also had lower rates of DNR code status on ICU discharge in this analysis.

We have highlighted what we believe are the most clinically relevant outcomes observed in this study. Although statistically significant differences were reached for some other outcomes including higher rates of mechanical ventilation for those with LEP and longer time on mechanical ventilation (0.3 days longer) this difference may not be very important in the context of the clinical environment.

This study has several limitations. First, this is a single center study from a tertiary care academic medical center in the midwest United States. Thus, generalizability to other settings may be limited. The proportion of those with LEP was lower than many centers around the US (2.8% v.8.5%) and the LEP cohort is heterogeneous and may be atypical in terms of country of origin.We used the widely accepted definition of Limited English Proficiency as a predictor variable however we recognise that some of the patients in the LEP cohort may have had excellent English language skills and proficiency despite English being their second language.The current definition of LEP used among federal agencies is not a perfect measure but it is commonly used in research and clinical practice,therefore while flawed is broadly recognised as a marker of potential difference and disparity^41–44. We were unable to prospectively measure individual authentic patient and family preferences, quality of decision making, or satisfaction. We relied on chart documentation of code status orders as an accurate representation of preferences. Evidence suggests that documented code status orders may not accurately reflect patient preferences in some instances.⁴⁵ Thus, we are not in a position to determine whether the differences we observed served the authentic values of the patients whose records were analyzed. Strengths of the study include, ability to abstract with high fidelity multiple pertinent outcomes from the electronic medical record, inclusion of multiple ICUs with variable practice settings, and an easily definable exposure measurement (LEP) that can be easily applied to other practice settings.

The sensitivity analysis offers some insights about potential confounders affecting the findings outlined above.The diversity within the LEP cohort incorporates a broad spectrum of patients with different religions, races and ethnicities and cultures whose beliefs about end of life may inform their decision making and care. However it should also be noted that the impact of having Limited English Proficiency may influence choices beyond what we commonly associate with cultural diversity and background^{46, 47}. Furthermore assessing health literacy and its impact on end of life decision making was beyond the scope of this work but is another vital element of effective decision making^48–50. Those patients with LEP may also have limited health literacy. We might however, expect health literacy to affect those with proficient English too.

Conclusions

The proportion of adults in the United States with limited English proficiency is approximately 8.5% and this proportion has increased over the past 20 years. Patients with limited English proficiency are at risk for adverse health outcomes. Even when adjusting for education level, and severity of illness, ICU patients with LEP are more likely to have lower rates of do-not-resuscitate orders, lower rates of comfort measure orders use before death, lower rates of advance directive completion, and higher rates of mechanical ventilator and restraints usage. When decisions to limit medical therapy are made, it took significantly longer to implement for patients with LEP compared to patients who spoke English well. Health care team members should be aware that these differences exist, and in conjunction with future research studies, should identify methods to improve life support decision making and end of life care for patients with language barriers. Healthcare system reforms designed to improve end of life decision making may need to devote extra attention to LEP populations. These reforms should include increased awareness of religious and cultural variations that may influence decision making and care at the end of life as well as consideration of health literacy issues. Language barriers however remain an independent risk factor for differences in decision making and care at the end of life potentially representing a disparity.

Supplementary Material

supplemental table

Sensitivity Analysis Supplementary Table: Outcomes in all patients with limited English proficiency compared to no limited English proficiency

Key Points.

Question:

Do differences exist in end of life care and decision making among patients with limited English proficiency compared to patients whose primary language is English?

Findings:

In this retrospective cohort study that included 27,523 patients, patients with limited English proficiency were less likely to have do-not-resuscitate code status, less likely to have advance directives, less likely to have a comfort measures only order prior to death, and were more likely to have mechanical ventilation and restraints used.

Meaning:

Patients with limited English proficiency have significant differences in end of life decision making, code status, and advanced directives compared to patients who speak English well.

List of abbreviations:

LEP

Limited English Proficiency

ICU

Intensive Care Unit

DNR

Do Not Resuscitate

APACHE III score

Acute Physiology and Chronic Health Evaluation

CPR

cardio-pulmonary resuscitation

IV

intravenous

EMR

electronic medical record\

OR

odds ratio

CI

confidence interval

IQR

interquartile range