Time Trends in Race-Ethnic Differences in Do-Not-Resuscitate Orders After Stroke

Abstract

Background and Purpose

Do-not-resuscitate (DNR) orders are common after stroke, though there are limited data on trends over time. We investigated time trends in DNR orders in a community with a large minority population.

Methods

Cases of ischemic stroke (IS) or intracerebral hemorrhage (ICH) were identified from the Brain Attack Surveillance in Corpus Christi (BASIC) study from June 2007 through October 2016. Cox proportional hazards models were used to assess time to DNR orders, with an interaction term added to allow separate hazard ratios for early (≤24 hours) and late (>24 hours) DNR. Stroke-type specific calendar trends were assessed with an interaction term between calendar year (linear) and stroke type.

Results

2,672 cases were included (ICH 14%). Mean age was 69, 50% were female, and race-ethnicity was Mexican American (MA, 58%), non-Hispanic White (NHW, 37%), and African American (AA, 5%). Overall, 16% had a DNR order during the hospitalization. For ICH, DNR orders (early and late) were stable over the study period. However, early DNR orders became more common over time after IS (hazard ratio for 2016 vs 2007: 1.89, 95% CI 1.06, 3.39), with no change over time for late DNR orders after IS. MAs (HR 0.65, 95% CI 0.50, 0.86) and AAs (HR 0.17, 95% CI 0.04, 0.71) were less likely than NHWs to have early DNR orders, though there were no race-ethnic differences in late DNR orders. There was no change in race-ethnic difference in DNR orders over the time of the study (interaction p > 0.60).

Conclusions

Despite revised national guidelines cautioning against early DNR orders in ICH, presence of DNR orders after ICH was stable between 2007 and 2016, with only slight increases in early DNR orders after IS. MAs and AAs remain less likely than NHWs to have early DNR orders after stroke.

Introduction

Do-not-resuscitate (DNR) orders are associated with increased risk of death after ischemic stroke (IS)¹ and intracerebral hemorrhage (ICH).² Specifically, early DNR orders (generally defined as within 24 hours of admission) have received particular attention in stroke in recent years due to the concern that early limitations in treatment may be premature given the uncertainties inherent in neurological prognostication.^{2, 3} The increased attention to early DNR orders resulted in the American Heart Association/American Stroke Association updating their guidelines for the management of ICH in 2007 to recommend avoiding new DNR orders in the first 24 hours after admission for ICH.⁴

Multiple studies have identified racial or ethnic differences in the prevalence of DNR orders after stroke, with Hispanic and African-American patients less likely to have DNR orders placed after ICH or IS than non-Hispanic Whites (NHWs).^5–7 It remains an open question whether these observed differences represent an accurate reflection of the preference of race-ethnic groups or a health disparity that needs to be addressed. Despite the consistent finding of race-ethnic differences in DNR orders, there has been limited study of calendar time trends in these differences.

Given increased attention to DNR orders in stroke, particularly in light of the 2007 update to national ICH guidelines,⁴ we investigated calendar time trends in DNR orders in patients with ICH and IS from 2007 to 2016. Our primary hypotheses were: 1) DNR orders would decrease over time, particularly early DNR orders for ICH; and 2) Minorities would be less likely to have DNR orders than non-Hispanic whites, but that this difference would decrease over time, perhaps due to less frequent DNR orders among NHWs over time.

Materials and Methods

The de-identified data that support the findings of this study are available from the corresponding author upon reasonable request from a qualified investigator as long as it conforms to HIPAA protection, IRB study approvals, and data use agreements that the study team has with the participating hospitals.

Study Population and Case Selection

Cases of IS or ICH were identified from the Brain Attack Surveillance in Corpus Christi (BASIC) study from June 2007 through October 2016 as previously described.^8–10 Stroke diagnosis was validated by study physicians in all cases based on source documentation blinded to ethnicity. Only the first stroke event (IS or ICH) during the study period for an individual was included, excluding individuals with missing data on DNR status.

The medical record was reviewed for key demographics, medical comorbidities, and insurance status (yes vs. no). Race-ethnicity was defined based on the medical record (supplemented by interview data if missing from the medical record). We have previously reported in this community that there is high agreement (kappa=0.94) between the medical record and self-report for race-ethnicity¹¹ and that the Hispanic population is predominantly Mexican American (MA).^{9, 12} Therefore, race-ethnicity was categorized as non-Hispanic white, MA, or African American (AA), with individuals of other race-ethnicity excluded from the analysis due to small numbers. National Institutes of Health Stroke Scale (NIHSS) was recorded from the medical record or abstracted from the chart if not documented.¹³ Mortality was determined from medical record review.

DNR status was determined by clear documentation of a DNR order in the chart. If available, date and time of DNR order was recorded. Time to DNR was calculated as time from stroke presentation to DNR order (in hours) and was censored at discharge, 30 days, or death, which ever came first. DNR orders were considered early if documented at or before 24 hours from admission, and late if documented more than 24 hours after admission.^{2, 14}

Statistical Analysis

Descriptive statistics for the study population were summarized with frequencies and percentages for the overall population as well as for patients with and without DNR orders. As a descriptive overview of DNR status, unadjusted probabilities of having a DNR order (both early and overall) were estimated with Kaplan-Meier survival estimates, stratified by stroke type, race-ethnicity, and calendar time period.

Cox proportional hazards models were used to investigate factors associated with time to DNR orders. To investigate calendar time trends, the number of years between the date of the stroke and January 1, 2007 was included as a predictor in the models (referred to as “calendar time” to distinguish terminology from “time to DNR”, which is the time from presentation to DNR). Other pre-specified model covariates included race-ethnicity (indicator variables for MA and AA with NHW as reference), sex, age quartiles, stroke type (ICH or IS), neurodegenerative disease (binary indicator for Alzheimer’s disease, Parkinson’s disease, amyotrophic lateral sclerosis, or dementia), heart disease (history of coronary artery disease, congestive heart failure, or myocardial infarction), cancer, chronic obstructive pulmonary disease, end-stage renal disease, cholesterol, hypertension, history of prior stroke, atrial fibrillation, diabetes, insurance status, and stroke severity (NIHSS quintiles). The proportional hazards assumption was checked using the cumulative sum of residuals test;¹⁵ this test revealed that stroke type, race-ethnicity (MA, NHW, or AA), calendar year, and neurodegenerative disease violated the assumption. In other words, it would not be appropriate to report a single time-constant hazard ratio for these variables as the effect changed based on time to DNR. To correct this violation of the proportional hazards assumption, binary interaction terms between DNR timing (≤24 hours, >24 hours) and each of those covariates were included in the model to treat their effects as time varying. That is, these interactions enabled us to estimate separate hazard ratios for these covariates for early (≤24 hours) and late DNR (>24 hours). This dichotomous representation of DNR timing for the interaction terms was selected based on visual inspection of the plots of the smoothed Schoenfeld residuals^{16, 17} and is consistent with prior definitions of early DNR orders after stroke.^{2, 14} It is worth noting that due to the nature of this binary DNR timing interaction term, estimates for late DNR are conditional on not having an early DNR. The linearity of the calendar year variable was checked by fitting separate models for early and late DNR and using the cumulative sum of residuals test.

To examine if calendar time trends differed by stroke type (ICH or IS) or race-ethnicity, other interaction terms were pre-specified and considered for inclusion in the model. A three-way interaction between stroke type, calendar year, and binary DNR timing (i.e. ≤24 hours, >24 hours) was investigated but removed due to lack of significance; two-way interaction terms for stroke type-DNR timing and stroke type-calendar year were retained. Similarly, a three-way interaction between race-ethnicity, calendar year, and DNR timing was investigated, though only the two-way interaction between race-ethnicity and DNR timing was retained due to lack of significance of the other interaction terms.

Two sensitivity analyses were conducted to address concerns about missing data for DNR status. One sensitivity analysis assess whether DNR status could be considered missing at random¹⁸ after adjustment for covariates and is described in the online Supplemental Material (please see http://stroke.ahajournals.org). A second sensitivity analysis assessed the stability of the conclusions on calendar time trends due to missing data in 2014 (no patients underwent medical record abstraction for a portion of 2014 due to a transition in study procedures). The model was re-fit limited to data through December 31, 2013 to compare the estimates for key covariates (stroke type over calendar time and race-ethnicity) to the full dataset to see if conclusions were similar.

IRB Approval

This project was approved by the University of Michigan IRB and the IRBs of the individual Corpus Christi hospitals. Written informed consent was obtained from included patients or a proxy who completed an interview; enrollment of non-interviewed participants was covered under an IRB approved waiver of consent.

Results

A total of 4967 cases of IS and ICH were identified by BASIC during the study period from June 2007 to October 2016. Of these, cases were excluded for recurrent stroke during the study period (534); ineligible race-ethnicity (60); DNR or discharge time recorded as before stroke (56); missing data time of stroke (38); missing data on DNR status (1592) or other covariates (15), leaving 2672 individual cases available for analysis. When comparing the 2672 included cases to the 1701 excluded for missing data, included individuals differed by race-ethnicity (percentage of the available individuals included by ethnicity was MA: 63%, AA: 62%, and NHW 58%, p=0.003) and were more likely to be younger (p=0.002), though there was no difference by sex (p=0.61) or stroke type (0.58).

Descriptive characteristics of the population by DNR status are shown in Table 1. The sample included 2295 (85.9%) cases of IS and 377 (14.1%) cases of ICH. Overall, 427/2672 (16%) had a DNR order during the hospitalization, while 2245 had no DNR order. Unadjusted Kaplan-Meier estimates of probabilities of having DNR (early and overall=early + late) are shown in Table 2, stratified by stroke type, calendar time period, and race-ethnicity. Consistent with prior literature, the probability of an early DNR order was higher in ICH than IS, with NHWs having the highest probability of early DNR and AAs the lowest.

Table 1.

Participant Characteristics

Category	All N = 2672	DNR Present N = 427	No DNR Present N = 2245
	N (%)	N (%)	N (%)
Stroke Type:
Ischemic (IS)	2295 (85.9%)	289 (67.7%)	2006 (89.4%)
Intracerebral Hemorrhage (ICH)	377 (14.1%)	138 (32.3%)	239 (10.6%)
Race-Ethnicity:
Non-Hispanic White (NHW)	981 (36.7%)	192 (45.0%)	789 (35.1%)
Mexican-American (MA)	1560 (58.4%)	221 (51.8%)	1339 (59.6%)
African American (AA)	131 (4.9%)	14 (3.3%)	117 (5.2%)
DNR Order Time Period
June 2007 to May 2010	650 (24.3%)	106 (24.8%)	544 (24.2%)
June 2010 to May 2013	907 (33.9%)	141 (33.0%)	766 (34.1%)
June 2013 to October 2016	1115 (41.7%)	180 (42.2%)	935 (41.6%)
Age
45 to 59	712 (26.6%)	46 (10.8%)	666 (29.7%)
60 to 69	664 (24.9%)	55 (12.9%)	609 (27.1%)
70 to 79	597 (22.3%)	101 (23.7%)	496 (22.1%)
80 and above	699 (26.2%)	225 (52.7%)	474 (21.1%)
Female	1341 (50.2%)	251 (58.8%)	1090 (48.6%)
History of Prior Stroke	702 (26.3%)	134 (31.4%)	568 (25.3%)
Neurodegenerative Disease*	314 (11.8%)	107 (25.1%)	207 (9.2%)
Heart Disease†	931 (34.8%)	195 (45.7%)	736 (32.8%)
Cancer	322 (12.1%)	69 (16.2%)	253 (11.3%)
Chronic Obstructive Pulmonary Disease (COPD)	298 (11.2%)	50 (11.7%)	248 (11.0%)
End Stage Renal Disease	138 (5.2%)	32 (7.5%)	106 (4.7%)
Hypercholesterolemia	1179 (44.1%)	156 (36.5%)	1023 (45.6%)
Hypertension	2188 (81.9%)	353 (82.7%)	1835 (81.7%)
Atrial Fibrillation	407 (15.2%)	118 (27.6%)	289 (12.9%)
Diabetes Mellitus	1212 (45.4%)	162 (37.9%)	1050 (46.8%)
Uninsured	367 (13.7%)	25 (5.9%)	342 (15.2%)
NIHSS
0 to 1	609 (22.8%)	35 (8.2%)	574 (25.6%)
2 to 3	527 (19.7%)	28 (6.6%)	499 (22.2%)
4 to 6	515 (19.3%)	47 (11.0%)	468 (20.8%)
7 to 12	476 (17.8%)	71 (16.6%)	405 (18.0%)
13 to 42	545 (20.4%)	246 (57.6%)	299 (13.3%)
Acute Ischemic Stroke Treatments‡
Intravenous tPA	241 (10.5%)	36 (12.5%)	205 (10.2%)
Intraarterial stroke treatment locally	5 (0.2%)	0 (0.0%)	5 (0.3%)
Transferred to out of area hospital	4 (0.2%)	0 (0.0%)	4 (0.2%)

^*Includes Alzheimer’s disease, Parkinson’s disease, amyotrophic lateral sclerosis, or dementia

^†Includes coronary artery disease, congestive heart failure, history of myocardial infarction

^‡Percentages for these items are only among ischemic stroke cases. Transfers from the emergency department to an out of area hospital are provided to estimate cases that may have been sent for consideration of thrombectomy.

Table 2.

Unadjusted Probability of DNR Orders (Kaplan-Meier Estimates)

Covariate	Early (≤24 hours) probability of DNR	Overall (early + late) probability of DNR
All cases	0.08	0.37
Stroke Type
Ischemic (IS)	0.06	0.30
Intra-cerebral Hemorrhage (ICH)	0.21	0.56
DNR Order Time Period:
June 2007 to May 2010	0.08	0.31
June 2010 to May 2013	0.08	0.43
June 2013 to October 2016	0.09	0.33
Race-Ethnicity
Non-Hispanic White (NHW)	0.11	0.37
Mexican-American (MA)	0.07	0.37
African-American (AA)	0.02	0.27

In the Cox proportional hazards model assessing time to DNR orders, patients with ICH were more likely than patients with IS to have a DNR order, both before and after 24 hours from admission (Table 3). The hazard ratios for the difference in DNR between ICH and IS were larger at the beginning of the study period than at the end (Table 3, “Stroke Type” rows). However, the reason for this difference was not due to reduction in early DNR orders in ICH over calendar time as we had originally hypothesized. Rather, early DNR after ICH remained stable over calendar time with no detectable change, while there was a slight increase in early DNR orders after IS (Table 3, “DNR Order Time Period” rows). The adjusted probability of DNR orders by stroke type over calendar year is shown in the Figure.

Table 3:

Adjusted Effects of Covariates on Time to DNR orders^*

Patient Characteristic	Early (≤24 hours) DNR		Late (>24 hours) DNR†
	Hazard Ratio (95% CI)	P-value	Hazard Ratio (95% CI)	P-value
Stroke Type
ICH vs. IS in 2007	4.42 (2.55, 7.63)	<0.001	3.38 (1.96, 5.84)	<0.001
ICH vs IS in 2016	2.39 (1.55, 3.68)	<0.001	1.83 (1.15, 2.92)	0.01
DNR Order Time Period
2016 vs. 2007 in ICH	1.02 (0.50, 2.10)	0.95	0.69 (0.32, 1.49)	0.35
2016 vs. 2007 in IS	1.89 (1.06, 3.39)	0.03	1.28 (0.73, 2.24)	0.39
Race-Ethnicity
MA vs. NHW	0.65 (0.50, 0.86)	0.003	0.84 (0.63, 1.13)	0.25
AA vs. NHW	0.17 (0.04, 0.71)	0.015	1.04 (0.56, 1.93)	0.90
MA vs. AA	3.75 (0.92, 15.26)	0.06	0.81 (0.44, 1.48)	0.50
Female Sex (vs. Male)‡	1.10 (0.90, 1.34)	0.37	1.10 (0.90, 1.34)	0.37
Age (reference 45 to 59)‡
60 to 69	1.4 (0.93, 2.11)	0.11	1.4 (0.93, 2.11)	0.11
70 to 79	2.47 (1.67, 3.66)	<.0001	2.47 (1.67, 3.66)	<.0001
80 and above	4.16 (2.85, 6.09)	<.0001	4.16 (2.85, 6.09)	<.0001
NIHSS (reference 0 to 1)‡
2 to 3	1.03 (0.62, 1.69)	0.92	1.03 (0.62, 1.69)	0.92
4 to 6	1.59 (1.02, 2.47)	0.040	1.59 (1.02, 2.47)	0.040
7 to 12	2.09 (1.39, 3.15)	<0.001	2.09 (1.39, 3.15)	<0.001
13 to 42	5.26 (3.67, 7.54)	<0.001	5.26 (3.67, 7.54)	<0.001

^*Also adjusted for insurance status, history of prior stroke, hypertension, atrial fibrillation, diabetes, cancer, chronic obstructive pulmonary disease, end-stage renal disease, high cholesterol, heart disease, neurodegenerative disease, and the interaction between neurodegenerative disease and binary DNR timing (early vs. late)

^†Estimates for late DNR (>24 hours) are conditional on not having had a DNR prior to 24 hours

^‡There was no interaction with DNR timing (early vs. late) for these covariates, therefore the estimates are reported as identical early and late DNR orders.

Figure:

Adjusted probability of early and overall DNR orders over calendar time for ischemic stroke and intracerebral hemorrhage. Note that the “overall” estimates were generated by summing the model estimates for early and late DNR categories to facilitate comparison with other literature. Estimates adjusted for sex, age category, race-ethnicity, insurance status, NIHSS, history of prior stroke, hypertension, atrial fibrillation, diabetes, cancer, chronic obstructive pulmonary disease, end-stage renal disease, high cholesterol, heart disease, neurodegenerative disease, and the interaction between neurodegenerative disease and time to DNR

Considering race-ethnic differences, MAs and AAs were less likely than NHWs to have early DNR orders, with no race-ethnic differences in late DNR orders (Table 3). There was no evidence of interaction between race-ethnicity and calendar year, suggesting that these differences in early DNR orders did not change over the study period (interaction P >0.60). There was no difference in DNR orders by sex.

The sensitivity analysis to address concerns about missing DNR status (Supplemental Material and Supplemental Table I, please see http://stroke.ahajournals.org) was consistent with DNR status being missing at random after adjustment for covariates. In the second sensitivity analysis restricted to data before 2014, there were no meaningful changes in magnitude or direction of effect of covariates related to race-ethnicity or in the yearly effect of stroke type on DNR status.

Discussion

Examining calendar time trends in DNR order after stroke, we found that early (≤24 hours) DNR orders after ICH were stable from June 2007 through October 2016 with a very slight increase noted for early DNR orders in IS. We observed no change in late DNR orders for either stroke type, though the point estimates for late DNR orders in ICH suggested a possible trend toward a decrease over the years of the study. Additionally, MA and AA patients were less likely than NHWs to have early DNR orders after stroke, with no evidence of change in the race-ethnic differences over calendar time.

We analyzed early and late DNR orders separately for several reasons. First, this separation was necessary to correct the violation of the proportional hazards assumption that would have otherwise occurred for our key covariates of interest. Second, early DNR orders have commonly been reported separately in ICH, with a typical definition of “early” DNR orders being within 24 hours of admission.^{2, 14} Our approach and selection of the 24 hour cut-point for early DNR allows comparison to other literature. We did not consider other definitions of the timing of early DNR orders, but this could be considered in future studies. Others have suggested that the timing of early versus late DNR orders may represent different clinical contexts.^{14, 19} Earlier DNR orders may be representative of prior clear patient preferences to limit intensive interventions, physician recommendations to limit treatment,²⁰ or perhaps extremes of initial severity or co-morbid illness. In contrast, later DNR orders could represent changes in the treatment plan after an initial trial of intensive treatment, worsening illness severity, complications, or changes in patient/family preferences over the course of the hospitalization. While we cannot determine the differences in context of DNR orders in the present study, it is important to keep these potential differences between early and late DNR orders in mind when interpreting our results.

The lack of change in early DNR after ICH over calendar time was in contrast to our original hypothesis, in light of the 2007 AHA ICH guidelines⁴ recommending against early DNR orders. It is possible that this finding reflects slow uptake of guideline recommendations or the belief that this guideline is controversial,^{21, 22} though we were not able to formally assess guideline adherence in the current manuscript. Prior work has shown that major barriers to guideline uptake also include lack of applicability to specific patients as well as guideline content clashing with patient preferences.²³ Although this study was conducted in single community, the overall prevalence of DNR orders we observed is similar to what has been reported in other US stroke populations.^{1, 2}

It is unclear what precisely underlies the increase in early DNR orders in IS in the study period. It is possible that it is part of the larger trend of the increasing physician comfort with counseling families on DNR orders,²⁴ thought this would not explain the difference we observed between ICH and IS. On the other hand, it is also possible that this increase reflects the changing preferences of the general public around end-of-life treatment with increased acceptance of palliative care^{25, 26} or changes in DNR documentation procedures over the course of the study. However, we were unable to assess changes in public attitudes or institutional or physician motivations regarding DNR documentation in the current study. It is unlikely that increasing use of thrombectomy for ischemic stroke affected our findings. Endovascular stroke treatment received a Class 1, Level A recommendation from the American Heart Association/American Stroke Association in late 2015, toward the end of our study period.²⁷ The hospitals in this community only recently developed the capability to perform thrombectomy locally, and less than 1% of the ischemic stroke cases had endovascular treatment. Finally, it is possible that the increase in early DNR in ischemic stroke is a spurious finding, and this should be confirmed in other datasets.

The race-ethnic differences in early DNR orders is consistent with prior literature describing race-ethnic differences in DNR among other medical populations, with AA generally being the least likely to have DNR orders and Hispanic populations intermediate between AA and NHW.^{6, 28, 29} These studies examining DNR orders have typically reported on Hispanic populations in general and did not separately report findings for MAs or other Hispanic subgroups, and it is uncertain if findings from other Hispanic groups apply to our population of predominantly second and third generation MAs. While we observed the largest differences between AAs and NHWs in early DNR orders, this finding should be interpreted with caution given the relatively lower number of AAs in our study. We had hypothesized that there may be a narrowing of the race-ethnic difference over the study period, possibly due to less frequent DNR orders in NHWs or increased DNR orders in other race-ethnic groups, though we did not identify any calendar time trends in these differences.

Reasons for the persistent race-ethnic difference in early DNR orders remain unclear and could be due to a combination of patient/family or physician factors. It is possible that the lower prevalence of DNR orders among the minorities accurately reflects differences in cultural or religious preferences for treatment.^{29, 30} Prior work has shown that MAs faced with situations involving neurological devastation with only a small chance of recovery more frequently wanted life sustaining technology compared to European-Americans.³¹ Among MAs, cultural emphasis on familial decision making and preferences around disclosure may differ from classical Western conceptions of advance directives for a single fully informed patient, which may also contribute to differences.³² Another possibility is that the greater mistrust of the medical system present among minority patients may translate itself into reduced acceptance of recommendations to limit interventions.³³ Considering provider factors, inability to effectively communicate poor prognosis to minority family members could play a role, though that seems less likely in this community where MAs are the majority.

Limitations

The data gathered for this study was done by trained abstractors reviewing medical charts where DNR status may be present in different areas of the chart. As such, there is some potential for misclassification of DNR status during chart review given the varying locations in which a DNR order may be documented. However, any misclassification of DNR status is likely to be random with respect to our key variables of interest and would therefore bias estimates toward the null hypothesis. Even in the early years of this study, the hospitals in this community generally documented DNR status on a separate signed and dated form in the chart, which should limit the possibility of changes in documentation quality over the study period. There is also the possibility of introduction of bias from missing data on DNR status or other covariates in 39% of otherwise eligible individuals in part due to changes in our chart abstraction procedures over the study period. However, sensitivity analysis suggested that DNR status was missing at random after covariate adjustment, which should limit these concerns. Assessment of patient-family preferences was not done during this study, though we are in the process of conducing additional interview studies with family decision makers about factors (including acculturation and religion) that may influence these decisions. We did not have detailed data on factors such as ICH volume, other life-sustaining treatments or interventions such as feeding tubes or brain surgery. Finally, this was a retrospective analysis of prospectively identified stroke cases conducted in a single community in South Texas with a prominent MA population and results may not be generalizable to other regions.

Summary

Despite revised AHA guidelines cautioning against early DNR orders in ICH, there was no change from 2007 to 2016 in DNR orders after ICH and a slight increase in early DNR orders after IS. Race-ethnic differences in early DNR orders were stable over the study period. Additional research into understanding reasons for the persistent race-ethnic differences in DNR orders after stroke is warranted.