Impact of Language Barriers on Stroke Care and Outcomes
Background and Purpose—Language barriers may lead to poor quality of care, particularly for conditions like acute stroke for which diagnosis and treatment decision making rely on taking an accurate patient history. The purpose of this study was to determine the impact of patient language barriers on quality of stroke care and clinical outcomes.
Methods—This retrospective cohort study used data from the Registry of the Canadian Stroke Network. All Ontario patients who were admitted with acute stroke or transient ischemic attack between July 2003 and March 2008 were selected. Mortality, stroke outcomes, in-hospital complications, quality of care, and disposition were compared between those without (n=12 787) and with (n=1506) language barriers, which was defined based on the patient’s preferred language. Hierarchical multivariable regression models determined the effect of language barriers, independent of baseline covariates.
Results—Patients with language barriers had better 7-day mortality than those without (7.0% versus 9.2%; OR, 0.69; 95% CI, 0.57–0.82; P<0.001). However, they were more likely to be discharged with a moderate-to-severe neurological deficit (65.9% versus 51.5%; OR, 1.25; 95% CI, 1.15–1.35). In-hospital complication rates did not differ, and quality of care indicators generally favored patients with language barriers.
Conclusions—Patients who had language barriers had reduced mortality and better performance on some quality of care measures. These differences existed despite adjustment for many potential confounders, including ethnicity, prognostic factors, and stroke characteristics.
Language barriers may lead to poor quality of care for many reasons, including misdiagnosis, delays in care, decreased patient engagement and empowerment, and misuse of diagnostic testing.1–3 However, for many medical conditions, clinical decision making can be based to a large extent on the result of objective physical examination findings, laboratory testing, or diagnostic imaging. Therefore, patients who do not speak the same language as their care providers need not necessarily receive poorer care, because making a diagnosis or selecting a treatment option can be done effectively despite the communication barriers. For example, language proficiency does not impact outcomes for diabetes mellitus or myocardial infarction.4,5 However, acute stroke is different. Making a diagnosis often relies on eliciting subjective patient symptoms, and treatment decisions, such as whether to administer thrombolysis, may hinge on definitively establishing the time since the onset of symptoms. Therefore, language barriers between patients and their healthcare providers may influence care and outcomes for acute stroke. For example, in previous research, longer lengths of stay and inadequate anticoagulation for such patients have been found.6,7
The objective of this study was to determine the impact of language barriers on the quality of care and clinical outcomes for acute stroke. We hypothesized that patients with language barriers would have lower quality of care and worse outcomes than those without language barriers.
Materials and Methods
Study Design and Data Sources
This cohort study used clinical data from the Registry of the Canadian Stroke Network (RCSN, now known as the Ontario Stroke Registry) from July 2003 to March 2008 in Ontario, Canada’s most populous province. The RCSN was used to prospectively collect data on consecutive patients with acute stroke or transient ischemic attack seen in emergency departments or hospitalized at 12 stroke centers in the province.8 Trained neurology research nurses collected data at each site using chart abstraction and care provider interview when needed. Detailed information was collected on patient demographics (including language preferences), presenting symptoms, risk factors, and other medical history, hospital care, interventions and complications, final disposition, discharge medications, and functional outcomes. Inter-rater reliability of chart abstraction for key variables was substantial to almost perfect (κ=0.66–1.0).
To collect longitudinal follow-up data on patients after their initial stroke hospitalization, the Registry was then linked with population-based healthcare administrative databases from the publicly funded health insurance program of the Ontario Ministry of Health and Long-Term Care, which provides coverage to all permanent residents of Ontario. These databases included the Registered Persons Database, which records demographic information including date of death for all residents; the Discharge Abstract Database, which records detailed diagnostic and procedural information for all hospital admissions; and the Ontario Health Insurance Plan’s Claims Database, which records all claims submitted by physicians for fee-for-service reimbursement. Individuals are linked between the Registry and all of these databases using a unique encrypted identification number.
Patient Population and Exposure Definition
All patients aged ≥20 years who were admitted to a Registry-participating hospital with an ischemic stroke, hemorrhagic stroke, or transient ischemic attack between July 1, 2003 and March 31, 2008 were selected. Patients who had a subarachnoid hemorrhage were excluded. For patients with multiple strokes included in the Registry, only the first event was included. Patients with missing data on key baseline characteristics (language preferences, socioeconomic status, and level of consciousness at presentation) were also excluded. Language barriers were defined based on the patient’s preferred language. Patients were considered to have no language barrier at most hospitals if their preferred language was English; at the Ottawa Hospital and the Sudbury Regional Hospital, which operate bilingually, patients whose preferred language was English or French were considered to have no language barrier. Patients who had other languages as their preferred language were considered to have a language barrier.
Outcome and Covariate Definitions
The primary outcome was early mortality (within 7 days) after stroke. Secondary outcomes were 30-day mortality, 1-year mortality, and (for patients who did not die in hospital) residual neurological deficit at discharge, measured as a modified Rankin scale between 3 (moderate disability) and 5 (severe disability).
To understand any potential observed differences in mortality, additional secondary clinical outcomes were evaluated: receipt of thrombolytic therapy, time from hospital arrival to thrombolytic therapy (door-to-needle time), in-hospital complications (decubitis ulcer, fall with fracture, pneumonia, and thromboembolic complications), length of stay, and discharge disposition from hospital (discharge to home or to a rehabilitation hospital versus transfer to another acute care institution or discharge to long-term care). Finally, a variety of quality of care indicators were studied, which were developed by a Canadian expert panel for inclusion in the Canadian Best Practice Recommendations for Stroke Care.9 These included brain imaging within 24 hours of hospital arrival; in-hospital assessment from an occupational therapist, physiotherapist, speech–language pathologist, social worker, and nutritionist; carotid imaging within 2 weeks of admission; and discharge prescriptions for aspirin or warfarin, antihypertensives, and statins. Detailed descriptions of each of these outcome variables and the eligible subset of the whole cohort in whom each was defined are described Table I in the online-only Data Supplement.
A variety of baseline characteristics were defined for each patient, including demographic, clinical, and stroke presentation factors. These are described in detail in Table II in the online-only Data Supplement.
The baseline characteristics were compared between patients with and without language barriers using χ2 tests for categorical variables and t tests for continuous variables. To determine the independent effect of language barriers on each of the primary and secondary outcomes, logistic regression models were built adjusting for age, sex, ethnicity, socioeconomic status, stroke type, level of consciousness on arrival, vascular risk factors, previous cardiovascular disease, previous cancer, dementia or depression, Charlson comorbidity score,10 and the variables included in the PLAN-IT clinical prediction rule, a validated tool to identify patients at risk for poor stroke outcomes.11 Length of stay was modeled similarly using quantile regression. The independent effect of language barriers on the in-hospital quality of care indicators was determined using logistic regression models adjusting for age, sex, ethnicity, socioeconomic status, stroke type, and level of consciousness on arrival only. Door-to-needle time was modeled similarly using quantile regression.
To explore in greater depth the factors that contribute to differences in the mortality outcomes between groups, we performed a series of logistic regression models: first unadjusted, and then sequentially adding the following covariates in a stepwise fashion: age and sex, then ethnicity and socioeconomic status, then stroke characteristics (stroke type and level of consciousness on arrival), then vascular risk factors and previous cardiovascular disease, then other comorbidities and PLAN-IT score, and then measures of aggressive versus supportive care during the admission (admission to a dedicated stroke unit, admission to an intensive care unit, use of a nasogastric tube, use of a permanent feeding tube, consultation with a neurologist, and palliative care).
To search for multicollinearity between the predictor variables, we examined the variance inflation factors for all the covariates when included together in a multiple regression model, and confirmed that they were all <5.12
All of the above multivariable regression models were constructed as clustered data models, adjusting for hospital site to account for the clustered nature of the data. Analyses were conducted using SAS version 9.3 (Cary, NC). Clustered quantile regression analyses were conducted using the lqmm package in R version 3.1.0.
The study was approved by the research ethics board of Sunnybrook Health Sciences Center, Toronto.
There were 17 080 admissions for stroke to a participating hospital among adults during the 5 years of the study. Those with subarachnoid hemorrhage (n=1555) were excluded, and the second or subsequent admissions for individual patients (n=741) were also excluded. Finally, 491 patients were excluded because of missing data on key variables (language preference, socioeconomic status, or level of consciousness on admission). Thus, the final cohort included 14 293 patients, of whom 1506 (10.5%) had a language barrier. The baseline characteristics of the included patients are shown in Table 1. Patients with a language barrier were older, were less likely to be white, were poorer, and had greater comorbidity than patients without language barriers. They were also more likely to present with an ischemic stroke and were less likely to present with a transient ischemic attack.
The 7-day mortality rate was 7.0% among those with language barriers and 9.2% among those without (P=0.006). When adjusted for baseline demographic and clinical differences, patients with language barriers were less likely to have died within 7 days after their stroke (adjusted odds ratio 0.69, 95% confidence interval 0.57–0.82, P<0.001). Similar results were seen for 30-day and 1-year mortality although the adjusted odds ratios became increasingly closer to unity with longer time periods (Table 2). In contrast, patients with language barriers were more likely to be discharged from hospital with a moderate-to-severe residual neurological deficit (65.9% versus 51.5%, adjusted odds ratio 1.25; 95% confidence interval, 1.15–1.35; P<0.001).
In-hospital complications were rare, and not statistically significantly different between those with and without language barriers. However, those with language barriers had longer median lengths of stay (1.6 days longer; 95% CI, 0.6–2.7; P=0.002). Although the crude proportion of patients who received brain imaging within 24 hours of admission was similar in both groups, those with language barriers were more likely to receive such imaging once age, sex, ethnicity, and socioeconomic status were adjusted for (adjusted odds ratio 1.30; 95% CI, 1.01–1.66; P=0.036). Patients with language barriers were more likely to receive assessments from a variety of healthcare professionals during the hospitalization. Thrombolysis, carotid imaging, and discharge prescription rates did not differ between groups.
To explore the factors that contributed to the observed mortality difference, we performed a series of regression models for the primary outcome, adjusting for additional covariates (Figure). In the unadjusted model, language barriers were associated with reduced 7-day mortality (odds ratio 0.82; 95% CI, 0.70–0.97; P=0.023). However, with sequential adjustment for demographic factors, then adding stroke characteristics, and then adding comorbidities, the adjusted odds ratio for mortality moved increasingly away from the null. However, adjustment for aggressive versus supportive care attenuated the mortality benefit for those with language barriers, with the 95% confidence interval nearly crossing unity. Similar results were seen for the 30-day and 1-year mortality outcomes (not shown).
Patients with language barriers had an ≈30% reduced odds of death at 7 and 30 days, and a >20% reduced odds of death at 1 year, even after adjusting for baseline demographic and clinical differences between groups. This reduction in mortality risk was attenuated by differences between groups in the desire for aggressive versus supportive medical care, and previous studies have suggested that early stroke mortality is strongly sensitive to patient/family preferences for withdrawal of care.13 As further evidence of this preference for active treatment, patients with language barriers were more likely to be discharged with a residual neurological deficit, had longer lengths of stay, and were more likely to be discharged to home or to a rehabilitation hospital rather than to another acute care or a long-term care institution. Patients with language barriers also performed better on several in-hospital quality of care markers: they were more likely to receive timely brain imaging, and were more likely to receive assessments from paramedical personnel. In fact, these differences in quality of care may have arisen because of language barriers: the healthcare team may have sought more intensive evaluation of patients with whom their communication was impeded. Alternatively, the longer lengths of stay for patients with language barriers may have provided more opportunities for paramedical assessments. Thrombolysis, carotid imaging, and appropriate discharge prescriptions were not associated with language barriers.
There has been little previous research examining the impact of language barriers on stroke care and outcomes. Differences in long-term (rather than in-hospital) mortality and quality of care have not been studied previously. A Canadian study examining the impact of language proficiency on lengths of hospital stay for various diagnoses found the greatest difference for stroke: 14.9 days for patients proficient in English versus 26.1 days for those who were not.6 These observations are similar to those in our study, though greater in magnitude. An American study of an anticoagulation clinic found that Hispanic patients, but not Asian patients, who were English speaking spent more time in therapeutic range than those who were not.7 Other studies did not find differences in stroke care or outcomes based on language barriers, but they may have been underpowered.14,15
Our study has several strengths to highlight, including a rigorous ascertainment of stroke care and outcomes using uniform definitions and trained abstractors. Rich clinical data were collected, allowing for evaluation of many clinically important measures of quality, and for extensive adjustment for potential confounding factors. Furthermore, we were able to link this rich clinical data with administrative data for long-term outcomes that, in the Canadian single-payer universal healthcare system, result in no loss to follow-up or missing data. In addition, the sample size in this study was markedly larger than virtually all previous studies, allowing greater power. However, there are some important limitations to note. Although we adjusted for a larger number of potential confounding variables in the analyses, as with any observational study, the associations that were found may be partially explained by residual confounding because of other unmeasured factors that differ between patients with and without language barriers. Second, notwithstanding the large number of participating hospitals, they remained predominantly urban tertiary care centers, so the generalizability of the results to smaller or less well-resourced centers is uncertain. However, the participating hospitals may actually reflect the areas where most patients with language barriers live because few immigrants settle in rural areas. Third, a language preference other than English (or French at the two bilingual hospitals) does not necessarily mean a patient would have language barriers: they may have sufficient fluency to be able to communicate with their providers, they may have family members or professional translators available to translate, or the healthcare providers may have been able to speak the patient’s language. However, many patients with even marginal English fluency may have self-reported an English language preference,16 which would bias the study toward the null.
Understanding the impact of language barriers on healthcare is important in many countries. For example, in Canada, the number of non-English–non-French speakers grew by >20% between 2001 and 2006, compared with only 4% growth in the number of English speakers.17 The reasons why patients with language barriers have lower stroke mortality, even after adjusting for prognostic factors, comorbidity, and stroke characteristics, remain unclear, though their greater desire for aggressive intervention over supportive care likely plays a role. Rather than being related to language barriers per se, the desire for aggressive care is probably related more to ethnicity, religious background, or immigration status, and further research is required to characterize which patients share this preference for aggressive medical intervention, and to understand the sociocultural factors explaining it.
In conclusion, although we hypothesized that patients with language barriers would have had poor outcomes after acute stroke because of difficulties with communication with the healthcare team, in fact we found they had reduced mortality, and better performance on some quality of care measures. These differences existed despite adjustment for many potential confounders, including ethnicity, prognostic factors, and stroke characteristics. Greater understanding is needed of the mechanisms through which these patients achieved better stroke outcomes, and of the methods of translating these mechanisms to the remainder of the population.
We wish to thank Eriola Asllani, Maria Chiu, Jiming Fang, and Jeremiah Hwee for their assistance with the acquisition of data and the analyses.
Sources of Funding
The study was funded by the Canadian Institute for Health Research (CIHR), grant number MOP-102641. Dr Shah is supported by a new investigator award from the CIHR. Dr Kapral is supported by a career investigator award from the Heart and Stroke Foundation, Ontario Provincial Office. The Institute for Clinical Evaluative Sciences (ICES) is a nonprofit research institute funded by the Ontario Ministry of Health and Long-Term Care (MOHLTC). The Ontario Stroke Registry is funded by the Canadian Stroke Network and the MOHLTC. The opinions, results, and conclusions reported in these studies are those of the authors and are independent from the funding sources. No endorsement by the funders, ICES, or the MOHLTC is intended or should be inferred.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.114.007929/-/DC1.
- Received October 28, 2014.
- Revision received December 22, 2014.
- Accepted January 6, 2015.
- © 2015 American Heart Association, Inc.
- Flores G,
- Laws MB,
- Mayo SJ,
- Zuckerman B,
- Abreu M,
- Medina L,
- et al
- Scheppers E,
- van Dongen E,
- Dekker J,
- Geertzen J,
- Dekker J
- Flores G
- Okrainec K,
- Booth GL,
- Hollands S,
- Bell CM
- 8.↵The Ontario Stroke Registry Web site. 2013. https://osr.ices.on.ca. Accessed October 15, 2014.
- Lindsay MP,
- Gubitz G,
- Bayley M,
- Phillips S
- O’Donnell MJ,
- Fang J,
- D’Uva C,
- Saposnik G,
- Gould L,
- McGrath E,
- et al
- Smith MA,
- Lisabeth LD,
- Bonikowski F,
- Morgenstern LB
- 17.↵Statistics Canada. Language Highlight Tables, 2006 Census. Ottawa: Statistics Canada; 2007. Report No.: 97-555-XWE2006002.