Language Proficiency: How the Physician Data Initiative’s Data Collection and Reporting Standard Impacts Graduate Medical Education Open Access

As the US population with non-English language preferences grows,¹  there is a need to improve health care access and communication quality across languages. Language concordance between patients and their physicians has been identified as the best method to improve outcomes and reduce language-related disparities for groups who prefer to communicate about their health in nondominant languages.^2,3  When a clinician’s skills are insufficient to directly communicate in the patient’s preferred language, working with medical interpreters improves care.^4,5  The effective use of a clinician’s language skills with patients and the recognition of when to request a qualified interpreter are practices that rely on clinicians being able to accurately self-assess their language skills.⁶  While imperfect, language self-assessment using a validated tool has been supported as beneficial for resource allocation (eg, identifying what departments/units need interpreter availability and in what languages)⁷  and for identifying which clinicians will most benefit from language courses⁸  or formal proficiency testing.⁹ 

For graduate medical education (GME) programs, identifying the language skills of applicants and current resident physicians can have multiple benefits. First, this data can help leaders identify learners and faculty members who are most prepared to provide language-concordant care for the local patient population.¹⁰  Second, language data can help leaders plan for the appropriate resources such that their residents are well-equipped to care for the linguistic groups prevalent at their institutions. Potential resources might include courses to help residents who would like to augment their skills in a language,¹¹  a formal certification examination to support residents who are already multilingual in confirming their skills,¹²  and hiring a sufficient number of qualified interpreters. Third, identifying multilingual learners may also point to potential strengths in communication skills, creativity, cognition, and problem-solving.^13,14 

Multilingual learners who gained their non-English language skills as a heritage language (ie, in their childhood home)¹⁵  often have personal family lived experience with language discordance in health care and may have themselves served as child interpreters.¹⁶  These lived experiences may uniquely prepare them to advocate for system-wide improvements and build trust with patients. Finally, multilingual learners may be motivated to care for patients with language preferences that align with their skills;¹⁷  practicing in areas where their skills are useful may provide meaning to their career and enhance well-being. However, without a clear understanding of learners’ languages and their level of proficiency in each, learners may be inappropriately taxed: Multilingual students and residents report frequently being asked to use their language skills in patient care even when they are unprepared to do so due to lack of training or limited language abilities.¹⁷ 

Some studies have reported on the language skills of physicians and learners.^10,18,19  Yet few organizations routinely collect data on physician language proficiency. Since 2013, the Association of American Medical Colleges (AAMC) has collected language data from applicants to the American Medical College Application Service (AMCAS) and the Electronic Residency Application Service (ERAS). The language proficiency question used in AMCAS and ERAS is a modification of a validated self-assessment tool called the Interagency Language Roundtable for Healthcare (ILR-H) scale.⁷  Despite language proficiency data being available to program directors who use the ERAS application, it is unclear whether and how program directors use it. To date, language data have not been collected by the Accreditation Council for Graduate Medical Education (ACGME) or by the American Medical Association (AMA). Recent AMA policy calls for the inclusion of language proficiency within the AMA’s Physician Professional Data.²⁰ 

To address gaps in physician workforce language data, the AAMC, ACGME, and AMA’s collaborative working group known as the Physician Data Initiative (PDI) sought to develop a shared standard on collecting and reporting language proficiency.

Representatives from each organization with expertise on language proficiency, policy, and data convened on a monthly basis from the spring of 2023 to the fall of 2024. Of the 3 organizations, only the AAMC had previously collected language proficiency data. The group analyzed AAMC’s language data collected through AMCAS and ERAS since 2013. Other resources included previously published scales for language proficiency self-assessment as well as the list of languages reported by the US Census and American Community Survey (ACS). After reviewing available sources, the PDI reviewed the AAMC’s previously used language question side-by-side with the ILR-H and proposed a draft standard. The PDI consulted with experts as needed and iteratively revised the draft until consensus was reached. In what follows, we highlight the main areas of PDI discussion.

The group determined that modifications were needed to rename the question as language proficiency rather than fluency, since the term fluency technically refers to the fluidity of speech, which is one of several dimensions of oral language production. The term proficiency better captures an individual’s overall skill level in the language and is more readily applicable to signed languages.

Drawing from both the AAMC’s language question and the ILR-H, the PDI selected proficiency level labels that maximized clarity. Uses of the term native were replaced with native/near-native to acknowledge that some individuals may have achieved the highest level of proficiency in a language even if they were not born into that language.^21,22  The 5 levels of proficiency on the PDI’s shared standard were labeled as follows: Native/near-native (equivalent to excellent on the ILR-H), advanced (very good on the ILR-H), good (unchanged from the ILR-H), fair (unchanged from the ILR-H), and basic (poor on the ILR-H).

Modifications to the category descriptions were also needed to update the wording for inclusiveness and clarity. For example, idioms (eg, “get the gist”) were replaced with clearer phrases (“get the general idea”). Also, the health care context was embedded throughout all levels, resulting in closer alignment with the ILR-H compared to the prior AAMC language question. The PDI standard diverged from the ILR-H in the use of the term “educated speaker,” which was not included in the recommended PDI standard since it incorrectly conflated the concepts of educational level with language proficiency.

Several sign language experts advised the PDI regarding how to best enable individuals to report skills in signed languages.²³  Consultants indicated that sign language proficiency is highly variable and context-dependent. In other words, an individual’s ability to effectively communicate depends not only on one’s own skills but also varies heavily depending on the skills of the person with whom one is communicating. While this is also true of spoken languages, it may be even more salient for signed languages because many individuals who are deaf or hard-of-hearing experienced a paucity of language exposure as children.²⁴  Even when proficient in American Sign Language (ASL), physicians may be unfamiliar with the more informal signing that patients may be using and may benefit from partnering with an interpreter. To make the PDI standard inclusive of signed languages, words that implied speaking were replaced with words that were inclusive of language production for both spoken and signed languages, such as conversing and communicating.

After drafting the standard, the PDI developed a list of language response options. The list was developed using the languages most commonly reported by respondents to AAMC’s ERAS application as well as the languages spoken by persons with limited English proficiency, as reported by the ACS. Spoken language names followed the ACS language code list,²⁵  which is also aligned with international coding standards.²⁶  A prominent area of PDI discussion involved the disaggregation of Chinese languages. While the US Census and ACS aggregate all Chinese languages, with Mandarin and Cantonese being the most prevalent,²⁷  consultation revealed that the distinction between Mandarin and Cantonese may be clinically substantial.²⁸  Hence, the PDI opted to include both languages as separate responses.

For signed language response options, ASL is the most common signed language in the United States. However, many countries have their own distinct signed language, and the use of lengua de señas mexicana, commonly known as LSM, is not uncommon in the United States.^29,30  Hence, a write-in option was provided to allow respondents to indicate proficiency in any additional signed languages separate from the write-in option for spoken languages.

In November 2024, the PDI released a language proficiency data collection and reporting standard and language list response options (Table).³¹  The standard consists of 5 proficiency levels and closely aligns with the ILR-H, but contains updated, more inclusive wording for enhanced clarity, inclusion, and usability. The language list includes 55 languages and 2 write-in options, one for spoken and one for signed languages.

The PDI’s language proficiency standard represents an important step in the recognition of multilingualism as an asset for physicians. Language skills have historically been an invisible demographic characteristic despite their demonstrated impact on patient outcomes.^2,32  The AAMC, ACGME, and AMA agreeing on a strategy for collecting and reporting physician language proficiency data will help advance the visibility of language in health care.

The language standard has several implications for GME (Box). First, language data can be useful for the recruitment of physicians motivated and well-prepared to communicate with linguistic groups prevalent in an institution’s local population. Second, language skills are not static and should be self-assessed at several time points in a physician’s career. For example, medical students who take a medical language course may increase their proficiency.⁸  Conversely, multilingual residents who match in a program where the non-English language they speak is not common may decline in skills.

Third, GME programs should note that self-assessed language data is an important step in characterizing learners and faculty members’ language skills, but self-assessment alone is not intended to certify proficiency for direct patient care. Prior research comparing the ILR-H to formal testing demonstrates the ILR-H’s accuracy at the highest and lowest levels.⁹  By contrast, physicians’ self-ratings in the middle of the scale are less accurate. Given its close alignment with the ILR-H, the PDI’s language standard is expected to perform similarly as a self-assessment tool and requires further study. Scholars have suggested that formal testing and certification is most appropriate for physicians self-assessing at advanced or higher language levels.^32,33  Those in good, fair, and basic levels may choose to enhance their language abilities such that they may eventually pass a formal test, but should be expected to work with qualified interpreters until then. The language proficiency standard has not yet been studied for signed languages, and, based on our expert consultation, self-reported skills in signed languages may be even less predictive of readiness for independent communication with patients than for spoken language skills. All learners, regardless of language proficiency, should be taught to effectively partner with qualified interpreters.^34,35 

Finally, it is important to consider potential unintended consequences of language proficiency data. For example, multilingual learners already report being pulled away from other responsibilities to serve as ad hoc interpreters for supervisors or peers.¹⁷  Collecting language proficiency data may further spotlight learners’ language skills and could inadvertently expose them to even more ad hoc interpreting requests. To guard against this, GME programs should have and disseminate clear policies that protect learners.³⁴  Hospitals may already have policies that prohibit the use of ad hoc staff or family members as interpreters,³⁶  as these practices violate legal requirements,³⁷  endanger patient safety,³⁸  and expose institutions to liability.³⁹  In reviewing existing policies or creating new ones, programs should partner with institutional language departments and qualified interpreters. GME leaders and faculty members should consider the role of language in evaluating learners. A subcompetency of the communication and interpersonal skills competency domain has been proposed to address gaps in evaluating residents’ skills in caring for populations with non-English language preferences.⁴⁰  Additionally, faculty members should recognize and avoid accent bias, which can negatively affect performance evaluation for learners with a perceived foreign accent.⁴¹ 

The newly proposed language proficiency standard should be evaluated in the context of GME programs and health care institutions and by seeking feedback from program directors and learners. Novel aspects of the standard, such as the disaggregation of Chinese languages and the explicit inclusion of signed languages, represent key opportunities for further study.

The authors thank all participants of the Physician Data Initiative and expert consultants for their contributions.

The ACGME News and Views section of JGME includes reports, initiatives, and perspectives from the ACGME and its review committees. This article was not reviewed through the formal JGME peer review process. The decision to publish this article was made by the ACGME.