Language as a variable

There are several challenges that arise when language is used as a quantitative variable in research. The first challenge is one of definition. Chu et al describe the heterogeneous approach to the measurement of language in the studies they reviewed as a limitation of their results. Some studies used dominant language proficiency, while others used preferred language, and yet others used primary language. Each measure assesses a different construct. And so, it becomes difficult to aggregate outcomes across studies when fundamentally different concepts are measured and compared.

Another challenge is that the borders around spoken language are assumed to be fixed and well defined. Thus, language is often made into a binary variable to facilitate comparisons between dominant language speakers and non-dominant language speakers. Yet, speakers of a non-dominant language may have some proficiency in the dominant language. Furthermore, the use of language is contextual. For example, a patient’s primary language may be Arabic, but they may prefer to receive their routine medical care in English, and mental health counselling in Arabic. Thus, while a binary variable facilitates statistical modelling and comparisons between groups, it also conceals the complexity of a patient’s linguistic practices.

Importantly, spoken language typically involves both speakers and listeners. However, the current literature almost exclusively examines patients’ language use and rarely reports or assesses that of clinicians. This makes the patient the focus of study rather than the clinician.2 Future research must consider how to measure the languages used by clinicians and their concordance with a patient’s preferred language, especially given the ample evidence that language concordant care is associated with improved patient outcomes.3–6

The role of dominant language speakers and institutions

Interactions between patients and clinicians occur within institutions that are embedded in a wider healthcare system and social context. A focus on the patient’s language use as the primary subject of study obscures the role of dominant language speakers, institutions and the health system in shaping linguistic equity. For example, do institutions offer meaningful access to interpretation services? Do they have policies that govern the use of interpretation? Do their hiring practices reflect the linguistic diversity of the communities they serve? What is the institutional culture as it relates to health equity? Understanding linguistic inequities requires that we move beyond measuring the patient’s language to assessing the degree to which institutions provide multilingual care.

Measuring the impact of interpretation services

Chu et al’s work underscores the challenges of understanding how interpretation services influence health outcomes such as readmission rates. In a stratified analysis by ‘access to or use of interpretation services’, the authors found higher odds of readmission from studies that did not describe interpretation services and no difference in readmission rates from studies that did describe interpretation services. Thus, the authors conclude that ‘interpretation access’ could be a mitigating factor in the association between language discordance and hospital readmissions. This finding should be interpreted with some caution due to the small sample size of the two studies analysed and the imprecise definition of interpretation access and use. One of the two studies verified patient-level interpretation and described the clinician who used the service but not the form of interpretation used (in-person, video, telephone).7 While the other study described the form of interpretation used (telephone) but was unable to match an interpreted encounter to a specific patient or describe the clinical context of the interaction.8

Thus, although the benefits of interpretation services are well established,9 a significant barrier to better characterising the relationship between interpretation and health system outcomes, such as hospital readmission, remains the absence of patient-level data on whether interpretation was received by a patient, what form it took (telephone, in-person, video), and what type of communication it was used for (medication reconciliation, discharge counselling, etc). Institutions must create policies to collect such data at the level of the individual patient while also taking steps to protect personal health information. This would allow for a more nuanced understanding of the causative factors underlying the association between interpretation and health outcomes of interest. It would also allow institutions to identify areas for quality improvement and move from simply providing access to interpretation services to setting benchmarks for interpretation use.

The need for theory to inform the claims we make

Chu et al’s work offers an opportunity to reconsider how we account for the relationship between a patient’s spoken language and a given health outcome. Typically, poor communication between language-discordant patients and clinicians is theorised to be the primary mediator of poor patient outcomes. Professional interpretation is known to improve clinical care.9 Thus, increasing the use of interpretation services is seen as the principal method for improving outcomes and remedying inequities for patients who speak non-dominant languages. However, the outcomes experienced by patients from language minority communities are likely shaped by more than poor communication alone, requiring an understanding of language as a social process.

Chu et al note that an important challenge in the analysis of research on linguistic inequities is the ‘complex relationship between language, race/ethnicity, migration status and socioeconomic status’. Language intersects with these characteristics in ways that are not well examined in conventional healthcare research. For example, it is unclear what proportion of the studies reviewed by the authors collected sociodemographic data from participants. Other studies that have explored these factors have found that patients who do not speak English in English-dominant countries are more likely to be racialised and live in poverty.10 11 Thus, when considering Chu et al’s findings, non-dominant language speakers may have less access to home care and other social services that support a transition from hospital to home. Using a sociolinguistic lens, we might also consider how perceptions of a person’s language use are linked to racism or xenophobia. For example, a white French speaker in an English-dominant institution may be cared for in a different manner from a French speaker who is not white.2 Similarly, language hierarchies position European languages above others,12 and discourses related to migration, assimilation, and visible markers of cultural or religious difference may also shape the care of speakers who are deemed to be inadequately proficient in the dominant language.

Moving forward

At a time when global migration is reshaping population demographics, Chu et al’s study provides an opportunity to reflect on the status of research on linguistic inequities in healthcare. First, it brings to light the need for a standardised approach to defining language use in order to better synthesise findings across studies. This requires greater coordination among researchers and the development of a consensus view on how best to report the complex dimensions of language. Health systems must also be mandated to collect such data. Self-reported language preference is a useful starting point,13 but we must develop the theory and methods needed to move beyond a binary understanding of language to one that considers the reality that patients and clinicians may speak multiple languages with varying proficiency and preference.

Second, the study reminds us to characterise the context in which communication occurs, so that the role of both speakers and listeners are examined. This requires that data on clinician language be included in health and administrative records so that language concordant care can be studied. It also underwscores the need for sociodemographic data be collected so that the scope of analysis can be expanded to include consideration of the effects of racism, xenophobia, religious discrimination and classism on health outcomes in non-dominant language speakers. Finally, the study highlights the need to develop measures and quality standards to assess how well institutions provide multilingual care. Such efforts must be grounded in the views of patients who speak non-dominant languages, and begin with collecting patient-level interpretation data.

Remedying linguistic inequities requires us to move from understanding language as a technical problem of poor communication,14 to one that accounts for social context. It requires us to enrich our view of language and its intersections with race, migration status, religion and socioeconomic position, and to recognise that interventions to address linguistic inequities should occur in concert with broader efforts to improve health equity. Only then can we better understand the associations between the variables we seek to describe and move towards substantive action to address the inequities experienced by language minority communities.