Patient-Physician Language Concordance, Antihypertensive Medications, and Cardiovascular Outcomes Among Allophone-Speaking Patients with Hypertension

Abstract

Background

Patient-physician language concordance is associated with better outcomes. However, the mechanism(s) explaining these associations are poorly understood. Our objective was to determine if antihypertensive medication use mediates the association between patient-physician language concordance and major adverse cardiovascular events (MACEs).

Methods

Our population-based, retrospective cohort study used data from the Canadian Community Health Survey (CCHS) from January 1, 2003 to December 31, 2014. We identified Allophone-speaking respondents (ie, the language spoken most often at home is one other than English, French, or an Indigenous language) with self-reported hypertension. We defined patient-physician language concordance as agreement between language spoken most often at home and language spoken with one’s regular medical doctor. Survey responses were linked to hospitalization and mortality records. We identified all MACEs within 5 years after survey completion. The associations between patient-physician language concordance, antihypertensive medication use, and MACEs were explored using multivariable logistic and Cox proportional hazards regression, respectively. The mediating effect of antihypertensive medication use was tested with natural effect models.

Results

We studied 5013 Allophone-speaking patients, including 1708 (34.1%) who received language-concordant care and 3305 (65.9%) who received language-discordant care. Patients who received language-concordant care were 38% less likely to experience a MACE compared to patients who received language-discordant care (hazard ratio 0.62, 95% confidence interval 0.48-0.80). No evidence was found that this association was mediated by antihypertensive medication use.

Conclusions

Patient-physician language concordance was associated with a lower risk of a MACE. However, this association was not mediated by antihypertensive medication use. Further research could explore potentially modifiable mediators of this association.

Hypertension is the leading modifiable risk factor for cardiovascular (CV) disease (CVD) worldwide.¹ With a prevalence of approximately 25% among adult Canadians,^2,3 hypertension is considered one of the most important risk factors for morbidity and mortality in Canada.⁴ Blood pressure control is effective at preventing adverse outcomes related to hypertension.^5,6 However, a recent study found that 35% of Canadians with hypertension have uncontrolled hypertension,⁷ a finding consistent with observations that national rates of blood pressure control have worsened over the past 2 decades.^8,9 A growing body of evidence shows that social determinants of health, generally defined as nonmedical factors that influence health outcomes,¹⁰ contribute to disparities in the management of hypertension and CVD,¹¹ and they may partially explain the stagnant trend in national rates of blood pressure control.

Canada’s linguistic diversity is increasing at an unprecedented rate.¹² At the time of the 2021 Census, nearly 1 in 4 Canadians had a mother tongue other than English or French, and 1 in 8 Canadians spoke predominantly a language other than English or French at home.¹² These 2 metrics, which identify Allophone-speaking residents (ie, those who primarily speak a language other than English or French),¹³ have both reached their highest proportions since linguistic information was first recorded in the 1901 Census.¹² As a result of this increase in linguistic diversity, many Canadians are now living in minority-language communities, a situation that occurs when individuals live in a region where their preferred or primary language is not spoken by the majority of the residents.¹⁴

The literature suggests that patients who live in minority-language communities often receive healthcare services with lower quality and safety levels, and they may experience worse outcomes compared to those of patients identifying with the majority-language group.15, 16, 17, 18, 19 In addition, patient-physician language concordance, which occurs when patients and physicians have proficiency in the same language, has been shown to be associated with better outcomes for patients with chronic diseases. For instance, observational studies conducted in the US have found that non-English-speaking patients with diabetes who received language-concordant care had better glycemic control compared to that of similar patients who received language-discordant care.20, 21, 22, 23 However, an important limitation of these studies is that they did not identify the mechanism(s) explaining the association between patient-physician language concordance and positive health outcomes. This knowledge gap represents an important barrier for the development and implementation of targeted interventions aimed at improving the health of minority-language communities.

We previously conducted a study to determine whether patient-physician language concordance was associated with a lower risk of CVD, hospitalization, and/or death over a 5-year follow-up period among a cohort of Canadians living with hypertension.²⁴ We found that Allophone-speaking patients who received language-concordant care had better outcomes compared to those of Allophone-speaking patients who received language-discordant care.²⁴ However, patient-physician language concordance was not associated with any of the outcomes for French-speaking patients.²⁴ The objectives of our study, which examined Allophone-speaking patients with hypertension living in Canada, were as follows: (i) to test the association between patient-physician language concordance and antihypertensive medication use among Allophone-speaking patients with hypertension living in Canada, and (ii) to determine whether the association between patient-physician language concordance and CV outcomes is mediated by antihypertensive medication use.

Material and Methods

Study design

We performed a retrospective cohort study using data from the Canadian Community Health Survey (CCHS).²⁵ The cohort consisted of Allophone-speaking respondents aged > 18 years with self-reported hypertension who were included in the CCHS from January 1, 2003 to December 31, 2014 and could be linked to hospitalization and vital statistics administrative data.²⁶ The complete inclusion and exclusion criteria are presented in Supplemental Appendix S1.

Data sources

The CCHS is an annual cross-sectional survey of the Canadian population that collects information on demographics, social determinants of health, and health status for approximately 65,000 respondents per year.²⁵ To identify respondents with hypertension, the interviewers first ask respondents the following question: “Do you have high blood pressure?”.²⁵ Since 2005, respondents who answered “no” or “don’t know” have also been asked the following question: “Have you ever been diagnosed with high blood pressure?”.²⁵ We considered all respondents who answered “yes” to either (or both) of these 2 questions to have a diagnosis of hypertension. Validation studies have shown previously that this definition has good agreement with the incidence of hypertension as collected from administrative data.^27,28

The CCHS has been linked to several administrative databases, including the Canadian Vital Statistics Database (CVSD) and the Discharge Abstract Database (DAD).²⁶ All respondents are asked to provide consent to have their survey responses linked to administrative databases. Respondents who did not provide consent (15.0%) were excluded. The remaining respondents were linked using probabilistic linkage (CVSD) and deterministic linkage (DAD). The linkage rate for the study cohort was 95.1% (Fig. 1).

Figure 1.

Study flow diagram. CCHS, Canadian Community Health Survey.

The CVSD collects information on all deaths in Canada, and the DAD collects information on all hospitalization in acute-care treatment facilities.²⁶ The DAD captures hospitalizations in all provinces and territories, except for Quebec.²⁹ Hospitalizations in Manitoba have been recorded only since April 1, 2004.²⁹ For this reason, we excluded all respondents living in Quebec, as well as respondents living in Manitoba before April 1, 2004.

Ethics

The microdata files from the CCHS are made available publicly through a mechanism set out by legislation and regulation. Given this, ethics approval is not required for their use (in accordance with the Tri-Council Policy Statement, article 2.2).^25,26 The study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines for observational studies.³⁰

Exposure

We identified Allophone-speaking respondents using their reported language spoken most often at home, which is a self-reported variable in the CCHS. Allophone languages include all languages other than English, French, and Indigenous languages (see Supplemental Appendix S2 for a complete list of the specific languages collected in the CCHS). Respondents who spoke multiple languages at home were excluded if 2 more languages corresponded to different linguistic groups (eg, Allophone-speaking and English and/or French).

All CCHS respondents are asked if they have a regular medical doctor. Since cycle 2 (2003-2004), those who report having a regular medical doctor are asked to provide the language that they usually speak with their regular medical doctor. For each individual respondent, patient-physician language concordance was defined as agreement between language spoken most often at home and language spoken with their regular medical doctor. Allophone-speaking respondents who spoke to their regular medical doctor in one of the Allophone languages that they spoke at home were said to have received language-concordant care. In addition, Allophone-speaking respondents who spoke to their regular medical doctor in a language coded as “other” were also considered to have received language-concordant care, as the “other” language was felt to represent an Allophone language not captured by the CCHS codebook. Allophone-speaking respondents who spoke to their regular medical doctor in another language (most commonly English and French) were said to have received language-discordant care. We excluded Allophone-speaking respondents without a regular medical doctor (n = 216; 4.1%).

Covariates

We obtained the following covariates from the CCHS: age (continuous); sex (female, male); marital status (single, married, or common-law); education (less than high school, high-school graduate, postsecondary graduate); household income quintile (1-5); geographic region (Western Canada, Eastern Canada); urban or rural residence (urban, rural); immigrant status (yes, no); knowledge of English (yes, no); history of diabetes (yes, no); history of heart disease; history of stroke (yes, no); obesity (body mass index < 25.0, 25.0-29.9, or ≥ 30.0); consumption of fruits and vegetables (< 5 times per day, 5-10 times per day, > 10 times per day); physical activity (inactive, moderately active, active); and smoking (current, former, never). A more detailed description of covariate measurement is provided in Supplemental Appendix S3.

Mediator

Antihypertensive medication use is a self-reported variable in the CCHS. Respondents with self-reported hypertension are asked if they have taken any medication for blood pressure in the last month and are provided with a binary answer choice (“yes” or “no”). In 2003 and 2004, this question was included in an optional module that was administered in only British Columbia, Quebec, and select health regions in Ontario. Since 2005, this question has been posed to all respondents with self-reported hypertension (regardless of province or territory of residence at the time of survey completion).

Outcomes

The primary outcome was incidence of a major adverse CV event (MACE) within 5 years of survey completion. A MACE was defined as the composite of hospitalization for acute coronary syndrome, hospitalization for heart failure, hospitalization for stroke, and CV death. The secondary outcomes consisted of individual components of the primary outcome, in addition to all-cause hospitalization and all-cause death. Outcomes were identified through diagnostic codes (Supplemental Appendix S4) that have been validated previously.^31,32

Statistical analysis

We reported baseline characteristics using means (with standard deviations) for continuous variables, and numbers (with percentages) for categorical variables. We compared baseline characteristics after stratifying by patient-physician language concordance/discordance, using t tests for continuous variables, and χ² tests for categorical variables. We compared outcomes after stratifying by patient-physician language concordance/discordance using the log-rank test.

The association between patient-physician language concordance and antihypertensive medication use was tested with multivariable logistic regression. The association between patient-physician language concordance and the primary and secondary outcomes was assessed with multivariable Cox proportional hazards regression. Models were adjusted for age, sex, marital status, education, household income quintile, geographic region, urban vs rural residence, immigrant status, knowledge of English, history of diabetes, history of heart disease, obesity, history of stroke, consumption of fruits and vegetables, physical activity, and smoking. Patients receiving language-discordant care were the reference group in all regression models. We used a cause-specific hazard model to adjust for the competing risk of death (or non-CV death, for the secondary outcome of CV death). We addressed missing data by performing multiple imputation by chained equations (Supplemental Appendix S5).

Adjusted odds ratios and hazard ratios (HRs) and their corresponding 95% confidence intervals (CIs) were estimated. Statistical tests were 2-tailed, and the significance threshold was set at P < 0.05.

Mediation analysis

We performed mediation analysis using natural effect models, which is based on the counterfactual framework.³³ The natural direct and indirect effects can be estimated with weighted regression analyses. First, the mediator (antihypertensive medication use) is regressed against the binary exposure (patient-physician language concordance) and the baseline covariates using an appropriate regression model. Given that the mediator (antihypertensive medication use) was a binary variable, we performed multivariable logistic regression while adjusting for the exposure and baseline covariates ($C$). Second, the dataset is extended by duplicating the original dataset, and a new variable, denoted $x^{\ast }$, is defined as 1–$x$ to represent the counterfactual of the binary exposure (patient-physician language concordance). Third, mediation weights, defined as $w_m=\frac{P\left(M=m|x^{\ast },c\right)}{P\left(M=m|x,c\right)}$, are calculated for each observation in the extended dataset. Fourth, the outcome is regressed against $x$ and $x^{\ast }$ using an appropriate regression model. Given that the outcomes in our study were time-to-event outcomes, we performed Cox proportional hazards regression—that is, $\lambda \left(y_{x,M_{x^{\ast }}}\right)={\lambda }_0\left(y\right)\mathit{exp}\left({\beta }_{1}x+{\beta }_2{x}^{\ast }+{\beta }_{3}c\right)$, where ${\lambda }_0\left(y\right)$ is the baseline hazard function. The natural and indirect effects are equal to ${\beta }_1$ and ${\beta }_2$, respectively. Robust standard errors were estimated using software for generalized estimating equations, and 95% CIs were defined as the point estimate of the natural direct and indirect effects +/- 1.96 times the robust standard error.³³

Results

We studied 5013 Allophone-speaking patients with self-reported hypertension who reported having a regular medical doctor (Fig. 1). The most common languages were Chinese (n = 754; 15.0%), Italian (n = 554; 11.1%), and German (n = 479; 9.6%). A complete list of individual Allophone languages (with corresponding frequencies) is presented in Supplemental Appendix S6. The mean age of the cohort was 64.3 years; 56.0% of respondents reported their sex as female. Overall, 1708 patients (34.1%) were receiving language-concordant care, and 3305 patients (65.9%) were receiving language-discordant care.

Baseline characteristics

Patients who received language-concordant care had similar distributions of age and sex compared to those of patients who received language-discordant care (Table 1). However, compared to patients who received language-discordant care, patients who received language-concordant care were more likely to be single, have a lower level of education, and have a lower household income. Patients who received language-concordant care lived almost exclusively in urban areas (98.2%), and a slightly smaller percentage of patients who received language-discordant care lived in urban areas (87.9%). The percentage of patients who reported having immigrated to Canada was higher among patients who received language-concordant care (96.8%) compared to the percentage who received language-discordant care (86.8%). Self-reported knowledge of English was lower among patients who received language-concordant care (57.3%), compared to the percentage who received language-discordant care (91.7%).

Table 1.

Baseline characteristics of Allophone-speaking respondents, stratified by patient-physician language concordance vs discordance.

Baseline characteristics	Allophones (N = 5013)
	Language-concordant care (N = 1708)	Language-discordant care (N = 3305)	P
Sociodemographic characteristics
Age, y, mean +/- SD	65.0 +/- 14.5	63.9 +/- 14.9	0.02
Sex			0.02
Female	1022 (59.8)	1787 (54.1)
Male	686 (40.2)	1518 (45.9)
Marital status			< 0.01
Single	557 (32.6)	942 (28.5)
Married or common-law	1150 (67.3)	2361 (71.4)
Missing	1 ( < 0.1)	2 ( < 0.1)
Education			< 0.01
Less than high school	739 (43.3)	1242 (37.6)
High school graduate	320 (18.7)	519 (15.7)
Postsecondary graduate	628 (36.8)	1496 (45.3)
Missing	21 (1.2)	48 (1.5)
Household income quintile∗			< 0.01
1 (lowest)	821 (48.1)	1173 (35.5)
2	313 (18.3)	767 (23.2)
3	195 (11.4)	452 (13.7)
4	120 (7.0)	325 (9.8)
5 (highest)	48 (2.8)	225 (6.8)
Missing	211 (12.4)∗	363 (11.0)∗
Geographic region			0.26
Western Canada†	719 (42.1)	1337 (40.5)
Eastern Canada‡	989 (57.9)	1968 (59.5)
Residence			< 0.01
Urban	1678 (98.2)	2904 (87.9)
Rural	30 (1.8)	401 (12.1)
Immigrant status			< 0.01
Yes	1653 (96.8)	2868 (86.8)
No	40 (2.3)	411 (12.4)
Missing	15 (0.9)	26 (0.8)
Knowledge of English§			< 0.01
Yes	978 (57.3)	3031 (91.7)
No	729 (42.7)	272 (8.2)
Missing	1 ( < 0.1)	2 ( < 0.1)
Comorbidities
History of diabetes	389 (22.8)	703 (21.3)	0.23
Missing	0 (0)	3 ( < 0.1)
History of heart disease	267 (15.6)	530 (16.0)	0.71
Missing	7 (0.4)	15 (0.5)
Obesity (by BMI)			< 0.01
Normal (BMI < 25.0)	706 (41.3)	982 (29.7)
Overweight (BMI 25.0–29.9)	559 (32.7)	1,315 (39.8)
Obesity (BMI ≥ 30.0)	258 (15.1)	777 (23.5)
Missing	185 (10.8)	231 (7.0)
History of stroke	81 (4.7)	166 (5.0)	0.65
Missing	1 (<0.1)	10 (0.3)
Health behaviours
Consumption of fruits and/or vegetables, times/d			< 0.01
< 5	922 (54.0)	1527 (46.2)
5–10	501 (29.3)	1211 (36.6)
> 10	39 (2.3)	89 (2.7)
Missing‖	246 (14.4)‖	478 (14.5)‖
Physical activity			0.48
Inactive	941 (55.1)	1836 (55.6)
Moderately active	332 (19.4)	702 (21.2)
Active	261 (15.3)	548 (16.6)
Missing	174 (10.2)	219 (6.6)
Smoking			< 0.01
Current	125 (7.3)	295 (8.9)
Former	424 (24.8)	1191 (36.0)
Never	1156 (67.7)	1809 (54.7)
Missing	3 (0.2)	10 (0.3)
Pharmacologic treatment
Antihypertensive medication use			0.04
Yes	1261 (73.8)	2339 (70.8)
No	268 (15.7)	589 (17.8)
Missing¶	179 (10.5)¶	377 (11.4)¶

Values are n (%), unless otherwise indicated.

BMI, body mass index; SD, standard deviation.

^∗Questions pertaining to household income were not asked to respondents living in the territories.

^†Western Canada includes British Columbia, Alberta, Saskatchewan, Manitoba, and the territories (Yukon, Northwest territories, and Nunavut).

^‡Eastern Canada includes Ontario, New Brunswick, Prince Edward Island, Nova Scotia, and Newfoundland. Quebec was excluded by study design.

^§Variable derived from question which asks respondents whether they speak English and French well enough to conduct a conversation in each language. Answer choices include: English only, French only, both English and French, and neither English nor French. Respondents who answered “English only” or “both English and French” were considered to have knowledge of English; those who answered “French only” or “neither English nor French” were considered to not have knowledge of English.

^‖Fruit and vegetable consumption was an optional module in cycle 3 (2005-2006).

^¶Antihypertensive medication use was an optional module in cycle 2 (2003-2004).

In terms of health behaviours, patients who received language-concordant care reported less-frequent consumption of fruits and vegetables and a less-frequent history of smoking compared to the frequencies for patients who received language-discordant care. No difference in physical activity occurred (Table 1).

Antihypertensive medication use

Of the 5013 Allophone-speaking patients included in this study, 3600 (71.8%) reported that they took an antihypertensive medication in the last month, and 857 (17.1%) reported that they did not take an antihypertensive medication in the last month; the remaining 556 respondents (11.1%) had missing values for this variable (Table 1). The percentage of individuals who reported taking an antihypertensive medication in the last month was slightly higher for patients who received language-concordant care, compared to the percentage who received language-discordant care (73.8% vs 70.8%, P = 0.04).

In the multivariable logistic regression analysis (Fig. 2), no statistically significant difference occurred in the odds of antihypertensive medication use for patients receiving language-concordant care vs patients receiving language-discordant care (odds ratio 1.11, 95% CI 0.91-1.35).

Figure 2.

Adjusted odds ratios for antihypertensive medication use and adjusted hazard ratios for major cardiovascular events (MACEs) in multivariable regression analyses. CI, confidence interval; MACE, major adverse cardiovascular event.

CV outcomes

Patients who received language-concordant care had fewer MACEs compared to the incidence among patients who received language-discordant care (6.5% vs 9.1%, P < 0.01). A statistically significant difference between these 2 groups also was observed for hospitalizations for acute coronary syndrome, hospitalizations for heart failure, and CV death (Table 2). However, no statistically significant difference occurred in the percentage of individuals who experienced hospitalization for stroke for patients who received language-concordant care vs patients who received language-discordant care (P = 0.72).

Table 2.

Outcomes experienced by Allophone-speaking respondents, stratified by patient-physician language concordance vs discordance

Outcomes	Allophones (N = 5013)
	Language-concordant care (N = 1708)	Language-discordant care (N = 3305)	P
Primary outcome—MACE∗	111 (6.5)	301 (9.1)	< 0.01
Secondary outcomes
Acute coronary syndrome	36 (2.1)	125 (3.8)	< 0.01
Heart failure	31 (1.8)	91 (2.8)	0.04
Stroke	40 (2.3)	71 (2.1)	0.72
Cardiovascular death	45 (2.6)	121 (3.7)	0.05
Hospitalization (all-cause)	527 (30.9)	1292 (39.1)	< 0.01
Death (all-cause)	135 (7.9)	322 (9.7)	0.03

Values are n (%), unless otherwise indicated.

MACE, major adverse cardiovascular event.

^∗Composite outcome of acute coronary syndrome, hospitalizations for heart failure, stroke, and cardiovascular death.

In the multivariable Cox regression analysis (Fig. 2), patients receiving language-concordant care were 38% less likely to experience a MACE compared to patients receiving language-discordant care (HR 0.62, 95% CI 0.48-0.80).

When considering secondary outcomes (Supplemental Appendix S7), we found that patient-physician language concordance was also associated with a lower risk of acute coronary syndrome (HR 0.60, 95% CI 0.39-0.90), heart failure (HR 0.51, 95% CI 0.32-0.83), CV death (HR 0.55, 95% CI 0.36-0.83), all-cause hospitalization (HR 0.72, 95% CI 0.64-0.81), and all-cause death (HR 0.84, 95% CI 0.50-0.81). Patient-physician language concordance was not associated with risk of stroke (HR 0.93, 95% CI 0.58-1.48).

Mediation analysis

Table 3 presents the parameter estimates for the total effect, the natural direct effect, and the natural indirect effect for each of the outcomes. The association between patient-physician language concordance and MACEs was not mediated by self-reported antihypertensive medication use (HR 1.00, 95% CI 1.00-1.00). The natural indirect effect also was not statistically significant for any of the secondary outcomes.

Table 3.

Total effect, natural direct effect, and natural indirect effect of the association between patient-physician language concordance and cardiovascular outcomes, mediated by antihypertensive medication use

Outcomes	Total effect∗	Natural direct effect	Natural indirect effect	% mediated by natural indirect effect†
Primary outcome—MACE‡	0.62 (0.48–0.80)	0.62 (0.48–0.80)	1.00 (1.00–1.00)	0.09
Secondary outcomes
Acute coronary syndrome	0.60 (0.40–0.89)	0.60 (0.40–0.89)	1.00 (0.99–1.00)	0.28
Heart failure	0.51 (0.32–0.83)	0.51 (0.32–0.83)	1.00 (1.00–1.01)	– 0.12
Stroke	0.93 (0.57–1.52)	0.93 (0.57–1.53)	1.00 (0.99–1.01)	0.20
Cardiovascular death	0.54 (0.36–0.83)	0.55 (0.36–0.83)	1.00 (0.99–1.00)	0.42
Hospitalization (all-cause)	0.72 (0.64–0.81)	0.72 (0.64–0.81)	1.00 (1.00–1.00)	0.44
Death (all-cause)	0.64 (0.50–0.81)	0.64 (0.50–0.81)	1.00 (1.00–1.00)	0.25

Values in parentheses are 95% confidence intervals.

MACE, major adverse cardiovascular event.

^∗Total effect = natural direct effect + natural indirect effect on log hazard scale (or natural direct effect x natural indirect effect on the hazard scale).

^†% mediated by natural indirect effect = 100 x natural indirect effect/total effect on log hazard scale (or 100 x log(natural indirect effect)/log(total effect) on the hazard scale).

^‡Composite outcome of acute coronary syndrome, hospitalizations for heart failure, stroke, and cardiovascular death.

Discussion

In this retrospective cohort study of Allophone-speaking patients with hypertension living in Canada, we found that patients who received language-concordant care had a 38% reduction in the relative risk of MACEs compared to the risk for patients who received language-discordant care. Allophone-speaking patients who received language-concordant care tended to have less-favourable social determinants of health (including lower education and lower household income) compared to those of Allophone-speaking patients who received language-discordant care. Despite these differences, patient-physician language concordance was associated with better outcomes in both unadjusted and adjusted analyses; our prior analyses also found no evidence of statistically significant heterogeneity in regard to the interaction between patient-physician language concordance and other social determinants of health for the primary outcome (MACE).²⁴ This finding highlights the role of patient-physician language concordance as an independent, favourable social determinant of health for minority language communities.

However, contrary to our hypothesis, no evidence indicated that the association between patient-physician language concordance and a lower risk of MACEs was mediated by antihypertensive medication use. Few studies have considered the association between patient-physician language concordance and use of cardioprotective medications. The existing literature, which is limited to Spanish-speaking patients with diabetes living in California, has provided inconclusive results. One study found no difference in the prevalence of use of antihypertensive or cholesterol-lowering medications after stratifying by patient-physician language concordance.³⁴ Two other studies found that medication adherence (among those who were prescribed antihyperglycemic medications) also was not associated with patient-physician language concordance.^35,36 However, a large population-based study assessed adherence to antihyperglycemic, antihypertensive, and cholesterol-lowering medications using dispensing data from pharmacies.³⁷ The authors of that study found no statistically significant association between patient-physician language concordance and adherence to any individual class of medications. However, patient-physician language concordance was found to be associated with adherence to all classes of medication when these were combined into a composite outcome.³⁷ Unfortunately, we could not control for medication adherence in our study, which used self-reported data obtained from the CCHS. A study of Ontarians aged > 65 years who were included in CCHS cycle 3 (2005-2006) found that antihypertensive medication use had only moderate agreement (kappa 0.55) and modest sensitivity and specificity (75% and 80%, respectively) compared to pharmacy dispensing records within 100 days of survey completion. If we assume that misclassification of antihypertensive medication use is nondifferential (ie, unrelated to patient-physician language concordance), then we would expect both the odds ratio in the mediator model and the natural indirect effect in the outcome model to be biased toward the null.³⁸ Although the possibility that enough misclassification occurred to explain the total effect in our analyses is unlikely, we believe that patient-physician language concordance may still be associated with small differences in the use of cardioprotective medications, due to differential medication adherence that could not be accounted for in our study. Future studies should quantify medication adherence; this could be done using pharmacy dispensing records (in administrative data), or alternatively with measurement of serum medication levels, medication event monitoring systems, or pill counting (in prospective clinical studies).^39,40

Given that the association between patient-physician language concordance and a lower risk of a MACE could not be attributed to antihypertensive medication use, further research is needed to identify potentially modifiable mediators of this association. Antihypertensive medications are integral for blood pressure reduction (which, in turn, decreases the risk of CVD), but several other interventions also are recommended for patients with hypertension. For instance, Hypertension Canada guidelines recommend salt restriction,⁴ as this has been shown to result in modest reductions in blood pressure,⁴¹ and possibly lower risks of MACEs and CV death.⁴² We hypothesize that patient-physician language discordance could be a barrier to the provision of culturally appropriate and effective dietary counselling. Unfortunately, we could not adjust for salt intake, as this variable was not collected in the CCHS. In addition, treatment of coexisting comorbidities (such as diabetes,⁴³ dyslipidemia,⁴⁴ heart failure,⁴⁵ and ischemic heart disease⁴⁶) also has been shown to decrease the risk of CVD independently of blood pressure reduction. In our cohort, these risk factors were prevalent (21.8% for diabetes, 15.9% for heart disease) and were strongly associated with CV events. Thus, one possibility is that some of the CV events observed during the follow-up period were attributable to suboptimal management of coexisting CV risk factors rather than hypertension alone. Although use of other cardioprotective medications (such as aspirin and cholesterol-lowering medications) is not available for CCHS respondents with hypertension, this information has been collected systematically for all CCHS respondents with diabetes since cycle 2 (2003-2004).²⁵ We are currently undertaking separate studies to test the association between patient-physician language concordance, cardioprotective medications, and CVD among patients with diabetes. We are hopeful that these studies will identify potentially modifiable mediators that are amenable to targeted interventions (eg, translation of patient education material regarding the benefits and risks of specific cardioprotective medications).

Limitations

Our study has limitations. First, a minority of respondents (15.0% of the overall cohort) did not provide consent to have their survey responses linked to administrative databases. This refusal could have introduced bias if the respondents who provided consent were systematically different from those who did not provide consent. However, our own analyses of unlinked CCHS files, which include respondents who did not provide consent to have their survey responses linked to administrative databases, found only slight differences for 2 of the covariates included in our study. Second, Allophone-speaking respondents were identified using the language spoken most often at home. Our interpretation of the results assumes that this language represents a respondent’s preferred language for healthcare encounters. Although this may not be true for all respondents, prior studies conducted among home care recipients and long-term care residents in Ontario, Canada found that the preferred language collected during resident interviews had strongest agreement with the language spoken most often at home (vs other language variables collected in the CCHS, such as first official language spoken, knowledge of English and/or French, and mother tongue).⁴⁷

Third, we excluded respondents who reported speaking multiple languages at home if 2 or more languages corresponded to different linguistic groups (eg, Allophone and either English and/or French). Thus, the results of this study may not be generalizable to respondents who speak an Allophone language at home in addition to either English or French. Fourth, some respondents may have received care from multiple physicians with different linguistic abilities; how these respondents would have answered the question asking them to specify the language that they speak with their regular medical doctor is unclear. Fifth, we could not control for the use of interpretation services. Although most jurisdictions in Canada have legislation that mandates the offering of professional interpretation services to patients who are unable to receive care in their preferred language, these services remain underutilized in practice.^48,49 Ad hoc or untrained interpreters do not consistently improve objective outcomes.^50,51 Given this finding, the possibility that patients who had interpretation provided by family members or friends had a significantly lower risk of CVD compared to the risk for patients who did not have access to any form of interpretation is unlikely.

Sixth, we did not have information regarding intensity of pharmacologic blood pressure treatment (ie, doses and/or number of antihypertensive medications) or blood pressure control achieved during the follow-up period. Next, our analyses may have been confounded by community-level factors (such as loneliness and social isolation), which have been shown previously to be associated with CVD.⁵² In addition, due to the cross-sectional nature of the data, the covariates, exposure, and mediator were collected at the same time. We cannot exclude the possibility of reverse causation (eg, antihypertensive medication use could have temporally preceded a change in the exposure or covariates). Lastly, the results of our study may not be generalizable to Allophone-speaking patients with hypertension living in the province of Quebec (who were excluded by study design).

Conclusions

In this retrospective cohort study of Allophone-speaking patients with hypertension living in Canada, patient-physician language concordance was associated with a lower risk of MACEs. However, contrary to our hypothesis, no evidence was found that this association was mediated by antihypertensive medication use. Future studies should explore whether the association between patient-physician language concordance and lower risk of MACEs can be attributed to other potentially modifiable evidence-based interventions that have been shown to decrease the risk of CVD.