Language discordance as a marker of disparities in cerebrovascular risk and stroke outcomes: A multi-center Canadian study

Highlights

• •Language discordance (when a patient's primary spoken language differs from the primary language of the health system) is associated with worse post-stroke functional outcomes and greater neurovascular risk compared to language concordant participants.
• •Language concordance is a simple, readily available marker to identify those at risk of worse outcome.
• •Adaptive care models, treatments and education strategies may be needed to mitigate barriers influenced by language discordance.

Abstract

Background

Differences in ischemic stroke outcomes occur in those with limited English proficiency. These health disparities might arise when a patient's spoken language is discordant from the primary language utilized by the health system. Language concordance is an understudied concept. We examined whether language concordance is associated with differences in vascular risk or post-stroke functional outcomes, depression, obstructive sleep apnea and cognitive impairment.

Methods

This was a multi-center observational cross-sectional cohort study. Patients with ischemic stroke/transient ischemic attack (TIA) were consecutively recruited across eight regional stroke centers in Ontario, Canada (2012 – 2018). Participants were language concordant (LC) if they spoke English as their native language, ESL if they used English as a second language, or language discordant (LD) if non-English speaking and requiring translation.

Results

8156 screened patients. 6,556 met inclusion criteria: 5067 LC, 1207 ESL and 282 LD. Compared to LC patients: (i) ESL had increased odds of diabetes (OR = 1.28, p = 0.002), dyslipidemia (OR = 1.20, p = 0.007), and hypertension (OR = 1.37, p<0.001) (ii) LD speaking patients had an increased odds of having dyslipidemia (OR = 1.35, p = 0.034), hypertension (OR = 1.37, p<0.001), and worse functional outcome (OR = 1.66, p<0.0001). ESL (OR = 1.88, p<0.0001) and LD (OR = 1.71, p<0.0001) patients were more likely to have lower cognitive scores. No associations were noted with obstructive sleep apnea (OSA) or depression.

Conclusions

Measuring language concordance in stroke/TIA reveals differences in neurovascular risk and functional outcome among patients with limited proficiency in the primary language of their health system. Lower cognitive scores must be interpreted with caution as they may be influenced by translation and/or greater vascular risk. Language concordance is a simple, readily available marker to identify those at risk of worse functional outcome. Stroke systems and practitioners must now study why these differences exist and devise adaptive care models, treatments and education strategies to mitigate barriers influenced by language discordance.

Introduction

Differences in ischemic stroke and transient ischemic attack (TIA) outcomes in primarily English-speaking health settings have been identified in Non-English speaking patients [1]. One large study demonstrated greater lengths of stay, stroke severity at discharge and prevalence of vascular risk factors in those with language barriers compared to those without [2]. Those whose first language is discordant with the dominant language of their health system may face unique barriers to high-quality health-care including greater readmissions to hospital, poorer understanding of hospital discharge instructions, and adverse events [3], [4], [5].

Language concordance is an understudied concept, and few studies have examined whether language discordance may influence ischemic stroke risk, outcomes, or comorbidity [6]. Some of the most prevalent post-stroke comorbidities are depression, obstructive sleep apnea (OSA) and cognitive impairment (DOC) [7]. The DOC co-morbidities predict long-term patient centered outcomes after stroke and may influence stroke rehabilitation and morbidity [7,8]. The DOC comorbidities are also risk factors for incident stroke [7,8]. To date, the DOC comorbidities remain underexplored, and there is a further need to identify factors that may modify their presence in the post-stroke setting.

In this multi-center cohort study across eight regional stroke centers in Ontario we assessed whether language discordance may predict differences in functional outcome, vascular risk and the DOC comorbidities in prospectively recruited patients with ischemic stroke or TIA.

Methods

This multi-center cohort study, the DOC Utility Study, took place between 2012 and 2018 at eight regional stroke centers in the province of Ontario, Canada. This study was independently approved by each institution's research ethics board. All ethics boards approved screening, data abstraction and analysis for this study with a waiver of consent as the study occurred in the context of routine care and was of minimal risk. The study met requirements of the Tri-Council Policy Statement: Ethical Conduct for Research Involving Humans (TCPS2). All centres had agreements to abstract data as part of the Ontario Stroke Network.

Population

Consecutive patients with a diagnosis of TIA or stroke, seen in a stroke prevention clinic, were eligible for inclusion in this study. Referrals to stroke prevention clinics came from emergency departments, inpatient stroke units, and primary care.

Exclusion criteria included: aphasia or motor dysfunction of sufficient severity that screening tests could not be completed, and non-fluent English without translation available. In addition, patients who had been admitted to long term care facilities at the time of clinic visit were excluded. Neuroimaging and clinical data were used to determine the diagnosis of stroke or TIA by a neurologist.

Exposure

On the basis of self-report at the time of the clinic visit, patients were Language Concordant (LC) if they were native English speakers, English as a second language (ESL) if they were fluent in English, but this was their second language, or Language Discordant (LD) if they were not English proficient and required translation.

Vascular risk factors

Clinical data relevant to stroke risk were abstracted from clinical charts, and included age, sex, vascular risk factors (smoking status, hypertension, type 2 diabetes, hyperlipidemia, atrial fibrillation), prior stroke/TIA, congestive heart failure, and myocardial infarction. Chart abstraction was done by the site coordinator, with training and auditing from the central coordinator, supported by the site and central primary investigators. We used standard data elements and definitions from the Ontario Stroke Registry, which are validated and reliable [9].

Outcomes

DOC screening

All patients were screened for Depression, Obstructive Sleep Apnea (OSA), and Cognitive impairment (DOC) using an integrated DOC screen (www.docscreen.ca) [10]. The DOC Screen is valid, feasible and reliable [11], drawing on elements of the Patient Health Questionnaire-2 for depression [12], the STOP questionnaire for OSA [13] and the abstraction, memory and clock drawing components of the Montreal Cognitive Assessment (MoCA) [14]. The DOC screening form includes the Modified Rankin Scale (mRS) which was used as a measure of functional outcome. After standardized training, the DOC screen was routinely administered by nurses, physicians, and research assistants in Stroke Prevention Clinics and urgent TIA/rapid-assessment clinics. For those who did not speak English, either a family member facilitated translation if they were present, or available staff members in the clinic who spoke the same language as the patient provided translation.

Previously published cut-offs were used to categorize patients as either high-risk or low-risk for depression, obstructive sleep apnea or cognitive impairment based on DOC screen scores [11,15]. High risk was defined as scores of ≥ 4 for depression, = 4 for OSA and ≤ 5 for cognitive impairment [11,15]. See Data Supplement for further information.

Statistical analysis

Kruskal-Wallis and Chi-squared tests were used to compare means and proportions for continuous and categorical variables of interest between LC, LD, and ESL cohorts.

Multiple binary logistic regression models were used to test the hypothesis that language concordance reveals differences in (i) depression (ii) obstructive sleep apnea and (iii) cognitive performance and (iv) functional outcome. Ordinal regression models were used to explore the relationship between language concordance and specific cognitive domains (memory, abstraction, clock drawing). Covariates entered into models included: age, sex, years of education, and mRS. In all ordinal logistic regression models the assumption of proportionality was fulfilled. All analyses were carried out using IBM SPSS Version 22.

Results

Population

A total of 8156 consecutive patients underwent DOC screening. Language concordance was not available for 150 (1.8%) patients, and patients with final diagnoses other than TIA and ischemic stroke were excluded (n = 1450), see Fig. 1. A total of 6556 patients (3499 ischemic stroke and 3057 TIA) were included in this analysis. 5067 (77.3%) were LC, 1207 (18.4%) ESL and 282 (4.3%) LD.

Fig. 1.

Patient Inclusion Flowchart.

Note: DOC: Depression, Obstructive Sleep Apnea and Cognition; TIA: Transient Ischemic Attack; CAD: Coronary Artery Disease; CT: Computerized Tomography; MRI: Magnetic Resonance Imaging

Vascular risk factors & depression, obstructive sleep apnea and cognition

We observed an increasing frequency of diabetes, dyslipidemia, and hypertension across our cohorts, with the lowest proportion observed in LC, greatest proportion observed in LD, and an intermediate proportion observed in the ESL cohort (Table 1). For between group differences in demographic and vascular risk factor data see Supplementary Table 1. Using logistic regression, we also examined whether language discordance predicts increased vascular risk. The results are reported in Supplemental Table 2 and overall we did note a greater odds of dyslipidemia and hypertension in LD and ESL cohorts while controlling for age, sex, education and mRS.

Table 1.

Demographic and clinical characteristics of patients with ischemic stroke or TIA by language concordance.

Variable N	Language Concordant 5067		ESL 1207		LD 282	p-value
Age (Years), Mean (SD)		67.2 (14.8)	67.4 (13.9)		74.4 (10.4)	<0.001*
Education (Years), Mean (SD)		13.9 (3.5)	13.3 (4.2)		10.8 (4.9)	<0.001*
mRS, Mean (SD)		1.1 (1.1)	1.1 (1.1)		1.6 (1.3)	<0.001*
Sex (Male), N (%)		2643 (52.2)	651 (53.9)		137 (48.6)	.235
Diagnosis (Stroke), N (%)		2682 (52.9)	646 (53.5)		171 (60.6)	.041*
Atrial Fibrillation, N (%)		764 (15.1)	191 (15.8)		56 (19.9)	.088
Diabetes, N (%)		1049 (20.7)	308 (25.5)		83 (29.4)	<0.001*
Dyslipidemia, N (%)		2722 (53.7)	706 (58.5)		191 (67.7)	<0.001*
Hypertension, N (%)		3171 (62.6)	839 (69.5)		229 (81.2)	<0.001*
Previous Stroke, N (%)		516 (10.2)	123 (10.2)		44 (15.6)	.014*
Previous TIA, N (%)		364 (7.2)	73 (6.0)		21 (7.4)	.362
Angina, N (%)		212 (4.2)	38 (3.1%)		11 (3.9%)	.254
Congestive Heart Failure, N (%)		145 (2.9)	40 (3.3)		8 (2.8)	.701
Prior CABG, N (%)		267 (5.3)	71 (5.9)		20 (7.1)	.328
Prior Myocardial Infarction, N (%)		437 (8.6)	89 (7.4)		30 (10.6)	.155
Valvular Heart Disease, N (%)		112 (2.2)	27 (2.2)		11 (3.9)	.180
Carotid Stenosis, N (%)		785 (15.5)	213 (17.6)		50 (17.7)	.133
Peripheral Vascular Disease, N (%)		238 (4.7)	58 (4.8)		11 (3.9)	.807
Smoking, N (%)		762 (15.7)		116 (10.1)	18 (6.9)	<0.001*

ESL = English as a Second Language; SD=Standard Deviation; CABG=Coronary Artery Bypass Graft; LC = Language concordant; LD = Language discordant; mRS =modified Rankin Scale; TIA= Transient Ischemic Attack.

The proportion of those classified as high-risk for OSA and depression using established cut-offs showed no significant differences between groups for OSA and depression (Table 2). As for cognitive testing, the percentage of patients who had low cognitive screening scores classified as high risk of cognitive impairment was significantly different between groups: LD 41.4%, ESL 27.5% and 16.5% in the LC cohort (X² = 162.6, p < 0.001; Table 2).

Table 2.

Depression, Obstructive Sleep Apnea and Cognition (DOC) scores in patients with ischemic stroke and TIA by language concordance.

Variable	Language Concordant	English Second Language	Language Discordant	Statistic
Depression (n = 6551)
DOC Mood
N	5063	1206	282
Median (IQR)	1.0 (1.6)	0.0 (2.0)	0.0(2.0)	H = 13.9, p = 0.001
High Risk (score ≥ 4), N (%)	492 (9.7)	141 (11.7)	35 (12.4)	X2 = 5.7, p = 0.057
Obstructive Sleep Apnea (n = 6547)
DOC Apnea
N	5060	1206	281
Median (IQR)	2.0 (1.0)	2.0 (1.0)	2.0 (2.0)	H = 7.1, p = 0.028
High Risk (score = 4), N (%)	317 (6.3)	62 (5.1)	21 (7.5)	X2 = 3.1, p = 0.213
Cognition (n = 6546)
DOC Cognition
N	5060	1206	280
Median (IQR)	8.0 (3.0)	7.0 (4.0)	6.0 (4.0)	H = 214.9 p < 0.001
High Risk (score ≤ 5), N (%)	837 (16.5)	332 (27.5)	116 (41.4)	X2 = 162.6, p < 0.001

SD=Standard Deviation; LC = Language concordant; LD = Language discordant; ESL = English as a Second Language.

Associations between language concordance, doc comorbidities and functional outcome

Depression, obstructive sleep apnea and cognition

The logistic regression models examining the relationships between language concordance and post-stroke DOC comorbidities are summarized in Table 3. No relationship was observed between language concordance and the risk of depression or sleep apnea. We did note a trend for increased risk of depression in the ESL (OR = 1.22, 95% CI= 1.00–1.50, p = 0.055) compared to the LC cohort.

Table 3.

Summary of logistic regression models examining the relationship between language concordance and risk of post-stroke comorbidities: depression, obstructive sleep apnea and cognitive scores.

DOC	Language	OR (95% CI)	p-value
Depression	Language Concordance		.150
	LD vs. LC	1.12 (0.76, 1.66)	. 574
	ESL vs. LC	1.22 (1.00, 1.50)	.055
Obstructive Sleep Apnea	Language Concordance		.299
	LD vs. LC	1.01 (0.62, 1.66)	.971
	ESL vs. LC	0.80 (0.60, 1.06)	.124
Cognition	Language Concordance		< 0.001
	LD vs. LC	1.71 (1.28, 2.30)	< 0.001
	ESL vs. LC	1.88 (1.59, 2.22)	<0.001

* All models reported are controlled for age, years of education, mRS score and sex;.

LC = Language concordant; LD = Language discordant; ESL = Second Language.

As depicted in Table 3, language discordance was significantly associated with lower cognitive screening scores. Given that cognitive scores could have been influenced by the effect of translation in the LD cohort, we examined group comparisons between (i) LC vs ESL cohorts and (ii) LC vs LD cohorts. Overall, both ESL (OR = 1.88, 95% CI= 1.59- 2.22, p < 0.001) and LD (OR = 1.71, 95% CI= 1.28- 2.30, p < 0.001) cohorts were more likely to have lower overall cognitive scores compared to the LC cohort. Older age, fewer years of education, higher mRS and male sex were also associated with lower cognitive screening scores. Language discordance was also associated with lower cognitive sub-domains: memory, abstraction and clock-drawing (Supplemental Table 2). Again, given the possible effect of translation in the LD cohort on cognitive sub-domains, between group comparisons were made between the LC and ESL cohorts. Overall, lower scores in all cognitive sub-domains were noted in the ESL cohort compared to the LC cohort (Supplementary Tables 3 – 5).

Functional outcome

We assessed the predictors of functional outcome as measured by the mRS. Poorer functional outcome was associated with increasing age (OR=1.005, 95% CI= 1.001 – 1.009, p = 0.005), fewer years of education (OR = 0.97, 95% CI= 0.958 – 0.983, p<0.0001), female gender (OR = 1.185, 95% CI= 1.08 – 1.30, p<0.0001), prior stroke (OR = 2.07 95% CI= 1.78 – 2.40, p<0.0001), atrial fibrillation (OR = 1.37, 95% CI= 1.20 – 1.56, p<0.0001), diabetes (OR = 1.28, 95% CI= 1.14 – 1.43, p<0.0001), and hypertension (OR=1.35, 95% CI=1.21 - 1.50, p<0.0001). Language discordance was associated with worse functional outcome as well, but only in the LD compared to the LC cohorts (OR = 1.66, CI= 1.33 – 2.08, p<0.0001) and not between LC and ESL speaking cohorts (OR=1.075, CI=0.956 – 1.21, p = 0.23).

Discussion

In this multi-center cohort study, we found that language concordance was associated with differences in demographic factors, cerebrovascular risk, post-stroke functional status and cognitive screening scores. LD and ESL participants were older and had fewer years of education. The average age of LD participants was 7 years older than the ESL and LC groups, which likely significantly contributed to their functional status and cognitive performance. Moreover, on average the LD cohort had less than 12 years of education while both ESL and LC had an average of 13 years; this suggests the LD participants likely had lower cognitive reserve, which could also affect their cognitive outcomes. Despite controlling for these demographic factors in our statistical models, these factors certainly play a role in degree of cerebrovascular risk, functional status and cognitive scores. This is consistent with prior studies which have described differences in vascular risk factors between English and Non-English speaking cohorts in English-speaking health care systems [1,16,17]. Differences in vascular risk could arise from other variables that were not captured by this study, including broader social determinants of health, ethnicity, risks or processes involved in immigration [18]. Even after controlling for covariates, including age and education, language discordant speaking patients in our study had 66% increased adjusted odds of having a poorer functional outcome. It is possible that such differences in functional outcome may reflect differences in stroke severity or delays in presentation after onset of stroke or the use of or access to preventative therapies or acute reperfusion therapies.

There are many interpretations of these results. Our primarily English-speaking health system in Ontario may be difficult to navigate for those who speak another language. Our results invite further exploration of how health systems and care models can be optimized for patients who speak another language. Furthermore, language concordance is likely a variable which measures many facets of an individual's social and cultural life. One's proficiency in the primary language spoken might reflect differences socioeconomic status or in access to education; or ethnocultural influences on health. In a cohort of 9881 internal medicine patients in Toronto, Ontario, a greater proportion of Non-English proficient patients belonged to the lowest income quintiles compared to English proficient patients [3]. In other studies, low socioeconomic status (as measured by education, neighborhood poverty, social class, household income, and occupation) has been associated with: an increased incidence and severity of stroke [19], reduced adherence to vascular risk-factor modifying medications [20], and lower survival rates after stroke [21]. One limitation of our study is that we only measured education as our main social determinant of health. Nonetheless, we propose that language discordance is an easily obtainable and pragmatic variable that can help practitioners identify potential patients who may have health care needs that require more time and attention to language barriers.

Language concordance may also reflect day-to-day language utility – that is – what it means to not speak the majority language of a society. In a primarily English speaking system, for instance, communication and social barriers created by the inability to speak English influences a patient's relationship with the health care system including their understanding of health care, adherence to medication, results in fewer primary care visits and willingness to attend follow up appointments [20,[22], [23], [24]].

Our study's findings with respect to the risk of lower cognitive screening scores warrants further investigation. While we observed lower cognitive scores in the language discordant cohort, the cognitive tests were administered in English. The use of English-based cognitive tests, reflecting English cultural biases and linguistic constructs, likely disadvantages a non-English speaking participant. To assess the possible influence of translation on cognitive scores we also compared our ESL and LC cohorts alone. Our ESL and LC cohorts were well matched with respect to demographic variables, and even after controlling for covariates, we found significant lower cognitive screening scores, including the sub-domains of abstraction, memory and clock-drawing. Even though our ESL speaking patients were fluent in English, we cannot exclude the possibility that a patient's ethnocultural or unique linguistic frameworks informed by their primary or preferred language could have influenced performance on a cognitive test administered in English. Overall, these findings have wide implications for every practitioner and researcher utilizing cognitive tests and raises important questions about how to address the influence of language on cognitive screening. Future studies should compare cognitive tests administered in English to cognitive tests administered in the patient's mother tongue to ascertain which cognitive tests may be most vulnerable to the effects of translation. Ideally, each patient should be assessed in their first language; however, this may be prohibitive in locations with many linguistic backgrounds. Ultimately, we must be cautious making conclusions about cognitive impairment in language discordant cohorts.

There are limitations of our study. Selection bias could have been introduced with the use of informal translators. It would be ideal for future studies to use formal translators, so that patients who do not have family or health-care providers readily available to help provide translation may be included, but this can be challenging in high-volume stroke clinic settings. We also did not utilize a formal cut-off criterion for English language competency and the quality and accuracy of translation may have differed between patients. Applying a formal criterion for competency in future studies may allow for more homogeneous groups. In addition, this study did not capture information about the specific language of each patient and their primary care provider - this too may be an important variable when considering how a patient with limited English proficiency may interact with a predominantly English-speaking healthcare system. Furthermore, different languages, such ideogram, morphogram and syllabogram based languages, utilize different neuroanatomic networks [25]. In this study we did not collect information about the different types of non-English languages. Given the differences in neural networks between different languages, in the setting of acute ischemic stroke, language type may be an important determinant of post-stroke cognitive performance. Future studies should examine language type as perhaps those languages that utilize broader networks may be more resilient to injury [25]. Future studies should also examine handedness, stroke lesion size, and location as potential determinants of depression, language, and cognitive function after ischemic stroke. With respect to our logistic regression models, we had a robust sample size and all models contained no more than 5 independent variables. There were no concerns for insufficient sample size in our models except for models examining the LD cohort with respect to depression and OSA, as the numbers of LD patients at high risk of depression and OSA was <50. Finally, patients with severe aphasia were excluded from the current study. Language discordance has potential to increase assessment bias on the part of clinicians diagnosing aphasia, especially in the acute phase when translation may not be available or planned in advance. In this study, based in clinics, patients had family members or translation available to reduce—but likely not eliminate—this bias.

There are several strengths of our pragmatic study worth noting. The consecutive recruitment of 6556 patients across multiple centers limits selection biases and improves the generalizability and external validity of this study. Also, patients were included without the large selection bias normally introduced by consented participation in stroke research, which, in many centers, often includes only English-speaking participants. This is one of the few studies to examine relevant post-stroke comorbidities (depression, sleep apnea and cognitive impairment) and functional status in a large cohort, including those who did not speak English. Finally, the large sample size provided unique power to assess even infrequent events and account for multiple variables.

In conclusion, this is the first multi-center, cohort study in a multi-cultural setting to suggest that assessing language concordance in patients with stroke/TIA may be helpful in revealing increased cerebrovascular risk and worse functional outcome. This study raises important considerations for health practitioners in multi-cultural settings where the language of an individual patient may be discordant from the region's primary language. In these settings, discordant language may help identify patients with unique health literacy and socioeconomic needs or may merely reflect a greater need to improve communication. This especially has important implications for post-discharge education for those patients who do not speak the primary language of the health system. Stroke systems and practitioners must design care models, education and assessments that anticipate and accommodate language discordance. Identifying language discordance early and mobilizing appropriate resources, may ultimately improve stroke outcomes. We invite future studies to examine whether language concordance may apply to other health systems, not just those that are predominantly English speaking. Further research is needed to determine how to reduce stroke risk and improve outcomes across diversity in languages, and how interventions for stroke prevention and rehabilitation can be tailored to patient specific linguistic and ethnocultural needs.

Sources of funding

This study was supported by an operating Grant-in-Aid from the Heart and Stroke Foundation of Canada Grant No. 000392, 2012–2014 and the Canadian Institutes of Health Research Grant No. 1012404. RHS has received salary support for research from Heart and Stroke Foundation Clinician-Scientist Phase II Award, Ontario Brain Institute, Sunnybrook and UofT Department of Medicine, Sandra Black center for Brain Resilience, Recovery and Repair. MKK holds the Lillian Love Chair in Women's Health at the University Health Network/University of Toronto.

Declaration of Competing Interest

None.