Alicia Fernandez, MD; E. Margaret Warton, MPH; Dean Schillinger, MD; Howard H. Moffet, MPH; Jenna Kruger, MPH; Nancy Adler, PhD; and Andrew J. Karter, PhD
As the number of Latinos with limited English proficiency and diabetes increases,1-3
health plans and physicians need to know if language barriers contribute to diabetes outcomes. Understanding the role of language barriers in health outcomes will assist in workforce planning (ie, hiring of bilingual clinicians) and contribute to quality improvement efforts.
Prior research has reported on the association of language barriers with glycemic control among patients with type 2 diabetes.4-6
Using data from an integrated health plan with language access services available, researchers found that rates of poor glycemic control (glycated hemoglobin >9%) were nearly twice as high among Latino patients compared with white patients, regardless of the Latino patients’ English language proficiency.4
Among limited English proficient (LEP) Latino patients, however, those with a language-discordant (ie, non–Spanish-speaking) physician were nearly twice as likely to have poor glycemic control compared with those who had a language-concordant physician. These differences persisted after adjustment for other demographic and clinical factors. A recent study in the same setting found a 10% increase in the proportion of LEP patients with good glycemic control among those who switched from a language-discordant primary care physician (PCP) to a language-concordant (Spanish-speaking) PCP compared with those who switched from a language-discordant PCP to another language-discordant PCP.7
Control of serum lipid levels and blood pressure is particularly important in diabetes, and over the last 10 years, diabetes care guidelines have underscored the need for monitoring and control of these powerful risk factors for cardiovascular and renal disease among patients with diabetes.8-10
Studies of Latinos with diabetes indicate a complex relationship among acculturation, diet, and exercise,11-14
with several studies finding that LEP Latinos have better lipid profiles than more acculturated English-speaking Latinos. Systolic blood pressure (SBP) may also vary with acculturation, although some study findings suggest no impact.11,15
In this study, however, we focused on the role of language barriers in healthcare and sought to determine the extent to which language barriers between physicians and patients may have impacted lipid and blood pressure control among LEP Latino patients with diabetes. By focusing on an insured population with uniform and continuous access to care, we are better able to isolate any contribution of language barriers to clinical outcomes.
This analysis used data from the Diabetes Study of Northern California (DISTANCE), an NIH-funded survey follow-up cohort study among members of the Kaiser Permanente Northern California (KPNC) Diabetes Registry.16
Conducted in 2005-2006, the DISTANCE survey included 184 questions designed to assess a wide range of social and behavioral factors that may influence diabetes-related outcomes. This survey was offered in multiple modes and languages to a race-stratified random sample of members of the KPNC Diabetes Registry. The survey had a response rate of 62%. DISTANCE was approved by the institutional review boards of the Kaiser Foundation Research Institute and the University of California, San Francisco.
KPNC is a nonprofit integrated healthcare delivery system providing comprehensive medical care to a diverse population of approximately 3.2 million members in Northern California. Distribution of patient demographic and socioeconomic factors is similar to that of the area population except in the extremes of the income distribution.17
Each KPNC facility provides bilingual clinicians and interpreter services through qualified bilingual staff, telephone language interpreters, and on-site professional interpreters for their LEP patients.
For this study, we used the same patient cohort as in the study of language and glycemic control by Fernandez et al.4
Study participants were DISTANCE respondents whose self-identified race/ethnicity was Latino (n = 3877) or white (n = 4521). We excluded those who had longer than a 60-day gap in health plan enrollment in the year prior to the survey date (n = 205), who did not have type 2 diabetes (n = 436), who did not respond to the question about their English language proficiency (n = 104), and who did not have either a low-density lipoprotein cholesterol (LDL-C) test or SBP measure (n = 294), leaving a total of 3463 Latinos and 3896 whites. For analysis of LDL-C, respondents who had no LDL-C measure (n = 683), had an abnormal liver function test (aspartate aminotransferase >150) (n = 34), or were missing physician information (n = 86) were also excluded from analysis. For analysis of SBP, respondents who had no SBP measure (n = 385), had end-stage renal disease or glomerular filtration rate less than 15 mL/min/1.73m2
(n = 228), or were missing physician information (n = 49) were also excluded from analysis.
Patient language status was assessed by the following DISTANCE survey question: “How often do you have difficulty understanding or speaking English?” Latino respondents who answered “usually” or “often” were designated as having limited English proficiency (“LEP Latino”), whereas those who responded “sometimes,” “rarely,” or “never” were designated as English-speaking (“English-speaking Latino”). Physician language ability was assessed by response to another DISTANCE survey question: “Without using an interpreter, how well does your personal physician speak your language?” Responses to this question were also dichotomized; participants who answered “well,” “very well,” or “excellent” were considered to have a language-concordant (ie, Spanish-speaking) PCP (LEP-concordant), whereas those who responded “fair,” “poorly,” or “does not speak my language” were considered to have a non–Spanish-speaking language-discordant PCP (LEP-discordant).4
LEP patients who did not respond to the question on physician language proficiency were excluded from the analysis of language concordance.
Other measures of interest determined from survey responses included demographic information and time since diabetes diagnosis. KPNC data from 2005 were used to generate individual comorbidity scores using the DxCg, a validated risk assessment tool designed to quantify a patient’s illness burden, in which higher numbers indicate greater illness burden.18
Patients whose benefit records showed no gap of greater than 1 month in pharmacy coverage in the year prior to the survey response date were categorized as having continuous pharmacy benefits. Patients were considered to be using antihypertensive medications or lipid-lowering agents if electronic records showed that they were dispensed any prescriptions on a list of commonly used agents in the year prior to the survey.
Our 2 main outcome measures were the patients’ most recent measurements of LDL-C and SBP obtained during routine clinical care in the year prior to the survey date. Poor lipid control was defined as LDL-C higher than 100 mg/dL and poor hypertension control as SBP higher than 140 mm Hg, using widely accepted clinical guidelines in place during study years. Secondary outcomes were mean LDL-C and SBP.
We compared characteristics of English-speaking white, English-speaking Latino, and LEP Latino patients using χ2
tests for categorical variables and Student’s t
tests for continuous variables. Among the LEP Latino patients, we compared language-discordant and -concordant groups. We compared the odds of poor lipid control and poor SBP control among the different patient groups using generalized estimating equations (GEE) models to account for covariates (age, sex, education, income, diabetes duration, comorbidities, continuous prescription benefits, and control medications) and clustering of patients by physician and healthcare facility. Because of sample size constraints, in the comparison of LEP language-discordant versus LEP language-concordant groups, we used a parsimonious GEE model accounting for clustering by facility and adjusting for age, sex, education, diabetes duration, missing comorbidity score, and control medications. We also specified models that included an interaction term to evaluate whether the LEP-LDL-C or LEP-SBP relationships differed by patient–physician language concordance. To determine if our results were sensitive to our definition of LEP, we repeated our analyses including patients who reported “sometimes” having difficulty with English in the LEP Latino group.
The study population (N = 7359) included 3896 English-speaking white, 2921 English-speaking Latino, and 542 LEP Latino patients. Clustering of LEP patients by physician was not common, as most (90%) of the LEP Latino patients were cared for by a Kaiser physician with 3 or fewer LEP patients in the study (range, 1-17).
English-speaking white, English-speaking Latino, and LEP Latino patients differed in several ways (Table 1
). Compared with either of the Latino patient groups, white patients were more likely to be male, report more education, and report greater income. When compared with their LEP counterparts, English-speaking Latino patients were more likely to be male (49.8% vs 36.7%; P
= .001), have finished high school (69.8% vs 28.2%; P
<.001), have annual incomes of $35,000 or above (56.6% vs 24.2%; P
<.001), and have continuous pharmacy benefits during the year prior to the survey (94.6% vs 86.4%; P
<.05). Mean comorbidity index values were similar among the 2 Latino patient groups and slightly lower than the mean comorbidity index for English-speaking white patients. Among the LEP patients, LEP-concordant and LEP-discordant groups had similar patient characteristics. English-speaking white patients were somewhat more likely to be dispensed lipid-lowering medications than English-speaking Latino patients (81.3% vs 76.1%; P
<.001), who were somewhat more likely to be dispensed these medications than LEP Latino patients (76.1% vs 71.2%; P
= .01). A similar pattern across patient groups was observed for antihypertensive medication dispensing.
Slightly more than one-third of all patients had poor LDL-C control, with a recent LDL-C level higher than 100 mg/dL. English-speaking Latino patients were somewhat more likely to have poorly controlled LDL-C than English-speaking white patients in unadjusted analysis (36.8% vs 33.7%; P
= .01) (Table 2
), whereas differences in lipid control between English-speaking Latino and LEP Latino patients were not statistically significant. We found no differences in lipid control among the LEP Latino patients by patient–physician language concordance. Similar patterns were observed for mean LDL-C by patient group.
Approximately one-fifth of patients had poor SBP control. The unadjusted percentage of patients with poor SBP control did not differ between English-speaking white patients and English-speaking Latino patients (21.7% vs 20.0%; P
= .11) or between English-speaking Latino patients and LEP Latino patients (20.0% vs 16.7%; P
= .08). Mean SBP was marginally lower (ie, better) in the LEP Latino group than the English-speaking Latino group (130.0 vs 131.4; P
Multivariate models adjusting for age and sex eliminated the small differences in lipid control between English-speaking white patients and English-speaking Latino patients, and further adjustment for other demographic and clinical factors did not change this finding (Table 3
) (adjusted odds ratio [AOR], 1.05; 95% CI,
0.93-1.18). However, among Latinos, LEP patients were less likely than English-speaking patients to have poorly controlled LDL-C (AOR, 0.71; 95% CI, 0.54-0.93). The interaction term for patient–physician language concordance by LEP status was not significant (P
>.05), indicating that the LEP-SBP and LEP-LDL-C relationship did not differ according to patient–physician language concordance. Analyses using mean LDL-C and SBP as outcomes in linear GEE models showed similar results. Sensitivity analyses including Latino patients who reported “sometimes” having difficulty with English (n = 598) in the LEP Latino group did not alter the findings (data not shown).
In a study of language barriers and lipid and blood pressure control among insured patients with diabetes receiving uniform access to care in an integrated health plan, we found no substantive differences in the prevalence of poor lipid control or poor blood pressure control among LEP Latino patients by the language ability of their physician. This contrasts markedly with the prior finding of large differences in poor glycemic control by ethnicity, language status, and language concordance in the same patient population.4
However, it is in line with the findings of a recent study reporting no improvement in lipid or blood pressure control when Latino patients with limited English proficiency switched from a language-discordant to a language-concordant PCP.7
We can only speculate as to why language barriers did not lead to differences in lipid or blood pressure control. First, about one-third of the patients in each group had poor lipid control despite high rates of dispensing of lipid-lowering medications, suggesting that difficulties in achieving patient adherence to lipid-lowering therapy may apply to patients regardless of their primary language.19-21
English-speaking Latinos had a somewhat higher prevalence of poor lipid control than LEP Latinos, even though LEP Latinos were less likely to have been dispensed medications for hyperlipidemia, suggesting a lower prevalence of high lipid levels among the LEP patients at baseline. In this context, limited English proficiency may function not only as a marker for language barriers, but also as a proxy for acculturation, which is known to adversely impact diet and lifestyle. Second, LDL-C and SBP control may be less sensitive to patient–physician communication than glycemic control. Lipid and blood pressure control are primarily functions of adherence to the appropriate medications; glycemic control often requires both medication and lifestyle changes. In another study from the DISTANCE cohort, Ratanawongsa et al reported better adherence for cardiometabolic therapies among patients who felt their doctors listened to them, involved them in decisions, and gained their trust; however, the communication–adherence association was stronger for glycemic control medications than for lipid and blood pressure medications.22
Clinicians in interpreted encounters often have difficulty eliciting patient viewpoints and values.23
Common misconceptions about insulin among LEP Latinos (eg, that insulin causes blindness or death)10,24
and discussions about patient lifestyle modification involved in glycemic control may be more easily addressed in language-concordant relationships than in interpreter-mediated clinical encounters, which tend to stay narrowly focused on symptoms and therapy.25,26
In short, counseling patients on lifestyle changes is likely easier and more effective in a language-concordant encounter than in one mediated through an interpreter.
Our study has several limitations. First, prior work has established that Kaiser Permanente patients with limited English proficiency are less likely to report problems with access to care or care quality than LEP patients in other major health plans,27
so our results may not generalize to other insured LEP Latino patients, particularly if their services are less robust. Robust language access services, such that clinicians and patients have easy access to interpreters, can greatly mitigate the impact of language barriers. Second, although it is possible that survey participants differ from nonrespondents, we found no substantive differences based on clinical data and Census-based socioeconomic data, which were available for all members of the cohort.16
Third, most Latinos in the DISTANCE cohort are of Mexican ancestry, so our results may not be generalizable across Latinos of other nationalities.28,29
Fourth, sample size limitations, particularly among LEP Latinos, could mean that we missed a small language-associated difference in outcomes, although this is unlikely to be of clinical significance. Fifth, we used exclusion criteria similar to those of an earlier study on glycemic control in order to create comparable populations; results might vary in other patient samples. In particular, we should note that patients who underwent no LDL-C or SBP testing were excluded from the study. These patients may be particularly vulnerable to language barriers, and their exclusion may bias our study toward the null.
We found no evidence that language barriers for LEP Latino patients with diabetes resulted in worse lipid or blood pressure control relative to English-speaking Latino patients or English-speaking white patients in a health system with access to interpreter services and low financial barriers to medications and clinical care. This contrasts with other research showing a strong association between language barriers and glycemic control and suggests that more research is needed to further our understanding of how patient–physician communication impacts healthcare outcomes. There is room for improvement in clinical outcomes among all patients with diabetes, including the growing Latino population. Effective strategies to improve lipid and blood pressure control for both English- and Spanish-speaking populations need to be developed, tested, and deployed.