Abstract
Background
Over 60 million patients in the USA have limited English proficiency (LEP) and experience barriers in care. Still, there exists no standardized method of monitoring the utilization of language interpreting services (LIS).
Objective
To introduce a methodological approach to systematically monitor utilization of LIS for LEP patients.
Design
We utilized a One-To-Many Match algorithm to align inpatient visits of LEP patients from the electronic health record (EHR) with corresponding calls from LIS billing logs, using a unique patient identifier (MRN) and LIS call dates within patient’s admit and discharge dates. Due to error when MRNs are recorded by LIS, the FuzzyWuzzy Probabilistic String-Matching technique was utilized to enhance match accuracy where exact matches were unattainable, addressing inherent complexities in language data matching.
Participants
The study involved 5823 inpatient encounters with a non-English preference in an urban hospital system in 2020, representing a linguistically diverse patient base, and attempted to match these against 183,655 LIS call logs.
Main Measures
Our approach successfully matched 83.1% (4389 out of 5823) of inpatient encounters to an LIS call.
Key Results
We observed significant language-specific disparities in LIS usage, with Spanish leading in call volume at 2737 calls (exact matches) and 845 (probabilistic matches). Concordance rates varied, exceeding 94% for all languages in exact matches and ranging from 53.9% for Arabic to 71.6% for Russian in probabilistic matches. The average frequency of LIS calls was about one call per day per language group in the inpatient setting.
Conclusions
The study provides vital insights into language service preferences, frequency, and duration. These findings emphasize the need for standard methods in monitoring LIS usage to enhance patient outcomes for LEP patients.
Supplementary Information
The online version contains supplementary material available at 10.1007/s11606-025-09378-w.
KEY WORDS: disparities, limited english proficiency, language interpreting service
INTRODUCTION
Healthcare disparities in the USA are deeply rooted in systemic inequities, often stemming from structural racism and discrimination (SRD). One stark manifestation of these disparities lies in the realm of communication, particularly affecting the over 60 million individuals who do not primarily speak English.1 Among them, 25.5 million people, constituting 9% of the US population are classified as limited English proficiency (LEP) patients.1–3 Effective communication forms the bedrock of quality healthcare, making it imperative to address the language barriers that LEP patients encounter during their medical journeys. These barriers manifest as irregular access to care, difficulties in contacting healthcare providers, delays in receiving timely care, and challenges in articulating questions to medical professionals.3,4 Alarmingly, healthcare organizations often resort to “unofficial” or “ad hoc” interpretation services, frequently relying on patients’ family members or friends for language assistance.5,6 Such practices engender a multitude of problems, often resulting in subpar care, adverse health outcomes, or even physical harm for LEP patients.4,7–10
US federal law mandates that healthcare providers offer interpreter services to these vulnerable populations, as stipulated in Title VI of the Civil Rights Act of 1964, the Americans with Disabilities Act, the Affordable Care Act, and the Health and Human Services (HHS) Culturally and Linguistically Appropriate Services (CLAS) Standards.11–15 However, little attention has been given to how a patient’s English proficiency is determined and how the need for interpretation services is conveyed to healthcare providers. This absence of standardized processes raises concerns about the ability of healthcare systems and providers to effectively coordinate care tailored to patients’ needs while reducing disparate outcomes.16 Additionally, variability across electronic health record (EHR) systems in documenting language interpreting services (LIS) usage may lead to data gaps in monitoring appropriate LIS usage for LEP patients.17,18 These gaps in knowledge underscore the urgent need for a systematic approach to address these issues.
The use of patient-reported data, including race, ethnicity, and preferred language, is crucial for identifying and addressing healthcare disparities.18 To address the challenges amplifying language barriers among limited English proficiency (LEP) populations, a robust method for objectively monitoring LIS compliance is essential. Historically, LEP patients were often treated as a homogeneous group in research, neglecting their diverse cultural and linguistic needs. Additionally, not all language barriers are alike —LIS are more available for some languages than others which puts certain language communities at further risk.5,6 Furthermore, studies oversimplified LIS usage, disregarding the nuanced reality of multiple LIS needs during patient visits, especially in inpatient settings. In this proof-of-concept study, we propose an objective measure of LIS usage by aligning call logs from the LIS vendor with medical encounters as a quality compliance metric. Our objective is to delineate the methods employed in this matching process and address the challenges encountered. We will commence with a discussion of direct matching, followed by an exploration of probabilistic matching.
METHODS
Study Design, Setting, and Patient Population
Our study is conducted within an urban healthcare system comprising community hospitals and tertiary academic centers. This healthcare system annually records 150,000 inpatient admissions and four million outpatient visits. Within this diverse patient population, limited English proficiency (LEP) individuals communicate in a staggering 228 different languages, accounting for approximately 9.8% of our patient population.
Medical Encounter Data (Data from Patient Registration Systems and EPIC®)
Demographic information for patient encounters is derived from various registration platforms and integrated into the EPIC® electronic health record (EHR) system. LEP patients were identified based on self-reported preferred language other than English during registration. For this proof-of-concept study, we focused on inpatient visits at our main academic center between January 1, 2020, and December 31, 2020, including visits requiring sign language interpreters.
Language Interpreting Service (LIS) Data (Data from Language Line Vendor)
Interpreting services within our health system can be delivered in person, telephonically, or via video remote interpreting (VRI). Due to the vast linguistic diversity in our patient base, most of our care utilizes phone or VRI services via our vendor, Language Line Interpreter. The vendor supplied detailed call logs for telephone and video calls requesting LIS during the study period. Interpreters manually record caller information, including the hospital employee identification number, patient’s medical record number (MRN), nature of the patient’s visit, needed language, and caller’s location within the hospital. Additional data, such as date, time, call duration, interpreter details, and call language, are automatically documented by the vendor’s system.
Data Analysis
Our primary objective was to align each inpatient visit (One) with corresponding LIS calls (Many) using a One-To-Many Match algorithm, as illustrated in Fig. 1. This analysis was rooted in gold-standard parameters including (1) LIS call dates, (2) LIS call site, (3) patient admit date, and (4) patient discharge date. Each patient encounter was uniquely identified by the MRN and Admit Date (Primary Key). We utilized Python 3, Polars v18.4 (Written in Rust) and JuPyter Notebook Integrated Development Environment (IDE) for Match Algorithms and SAS v9.4 for statistics analysis.
Figure 1.
Matching algorithm specifically between hospital patient encounters and LIS calls.
Matching Variable Preparation
MRN Variable—MRNs adhered to institutional standards, devoid of special characters, and were either purely numeric or started with a letter followed by a number. MRNs spanned four to eight characters (e.g., E1234, 1234567).
LIS Call Data—We focused on the top five languages spoken at the main hospital: Spanish, Chinese, Russian, Bengali, and Arabic. Other languages were categorized as “Other,” with sign language treated as an additional category. Call durations and costs for identical call dates were aggregated (e.g., if a patient had one call on 1/1/2020 lasting 10 min and another call on the same date lasting 2 min, we collapsed this to get a duration of 12 min on 1/1/2020). This provides us with a better idea of the duration of interpretation services, which is an appropriate metric for assessing the quantity of services. If more than one language was listed for the same MRN, the was determined by whichever call had the maximum duration. There are occasional instances where the wrong language may be mistakenly entered for a patient. This can happen due to the complexities of managing multiple patients, the diversity of languages spoken, and the fast-paced environment in which interpreters often work. Such errors are typically unintentional and can be quickly rectified once identified. Recording the language with the longest duration gives the best estimate for the correct language, in case another one has been mistakenly recorded in another entry. We encountered no such instances in our dataset.
Patient Encounter Data— Patient visits were differentiated by MRN and admit date. In instances of overlapping admit dates, the most recent discharge date was retained. For example, patient with MRN 1234 (admit date=7/24, discharge date=7/28) was retained, while patient with MRN 1234 (admit date=7/24, discharge date=7/25) was removed. This is particularly important when dealing with situations where the same patient might have multiple hospital charts for the same visit. By focusing on the most recent discharge date, we effectively avoid including redundant data, ensuring that each visit is represented accurately in the dataset. We encountered one instance of this in our data set.
Matching Method Phase I—Exact Match by MRN and LIS Call Date
We conducted an exact MRN match between the Main Hospital inpatient encounters and the LIS call log. A match was considered exact if the LIS MRN perfectly aligned with the visit MRN, and the LIS call date fell within the admit and discharge dates. successful matches underwent summary statistics calculation, and the corresponding call logs were excluded from future match algorithms. Since a single patient visit could include multiple LIS calls, only LIS call data was removed for an exact match, while all patient encounter data was retained for probabilistic matching (Fig. 1). Once an exact match was confirmed in the call data, we removed that MRN from the call log to prevent reusing that call data again in our match algorithm.
Matching Method Phase II—FuzzyWuzzy Probabilistic String Matching
Interpreters often made errors when recording MRNs in the LIS call data, leading to discrepancies that an exact MRN match might not capture. To address this and account for these discrepancies, we employed FuzzyWuzzy, an open-source fuzzy string matching tool developed by SeatGeek, to conduct a probabilistic match on all main hospital inpatient encounter data.19,20 The algorithm aimed for optimal Levenshtein distance scores, targeting approximately 80–85%, which aims to strike a balance between identifying true matches and avoiding false positives.21,22 Similar to Matching Method Phase I above, the criteria for a match required the LIS call date to fall within the Admit to Discharge date range, and the call location to align with the patient’s location during their hospital stay. Since this method did not solely rely on MRN matching, it was capable of rectifying some LIS call data errors, a prime strength of probabilistic matching when certain data fields might be missing or incorrectly inputted. A match was deemed successful if there was consistency in location between the patient encounter and the LIS call log, and if the call date occurred within the patient’s admission period (Fig. 1). To account for the number of possible recording errors, the number of probabilistic matches gives a good estimate, as an exact match was not able to be secured.
RESULTS
Our analysis encompassed 50,424 inpatient encounters from the calendar year 2020 at the main hospital. Among these, 5823 encounters indicated a non-English preferred language, including 30 visits that utilized a video American sign language interpreter. During the same period, the LIS call logs for the main hospital registered a total of 183,655 calls. Table 1 displays the demographics of these encounters. The majority of LEP patients self-identified as Hispanic (n=3226, 55.4%), followed by Asian (n=912, 15.7%), White (n=838, 14.4%), Other (n=692, 11.9%), and Black (n=155, 2.7%). Out of the 5823 inpatient encounters with a non-English preference, our objective was to align them with the 183,655 MSH LIS call logs from 2020. Specifically, 3134 inpatient encounters (53.8%) resulted in successful exact matches, 360 (6.18%) yielded at least one probabilistic match, and 895 inpatient encounters (15.4%) had both exact and probabilistic matches. In other words, 4029 (69.2%) patients encounters were able to be matched “exactly” through either only exact match or with exact match and probabilistic match. Therefore, a total of 83.1% (4389 out of 5823) of these patient encounters were effectively matched to at least one LIS call using either the exact or probabilistic match methods (Table 2).
Table 1.
No. of Encounter
| English speaking | LEP | Total | p-value* | |
|---|---|---|---|---|
| Age(years) mean(+SD) | 44.02 (27.38) | 53.77 (28.58) | 45.15 (27.70) | <0.0001 |
| Race, n (column %) | <0.0001 | |||
| Asian | 2524 (5.66%) | 912 (15.66%) | 3436 (6.81%) | |
| Black | 9392 (21.06%) | 155 (2.66%) | 9547 (18.93%) | |
| Hispanic | 7930 (17.78%) | 3226 (55.4%) | 11,156 (22.12%) | |
| Other | 6140 (13.77%) | 692 (11.88%) | 6832 (13.55%) | |
| White | 18,615 (41.74%) | 838 (14.39%) | 19,453 (38.58%) | |
| Sex, n (column %) | <0.0001 | |||
| Female | 24,758 (55.51%) | 2992 (51.38%) | 27,750 (55.03%) | |
| Male | 19,843 (44.49%) | 2831 (48.62%) | 22,674 (44.97%) | |
| Insurance, n (column %) | <0.0001 | |||
| Commercial | 17,616 (39.5%) | 1100 (18.89%) | 18,716 (37.12%) | |
| Medicaid | 11,692 (26.21%) | 1910 (32.8%) | 13,602 (26.98%) | |
| Medicare | 12,830 (28.77%) | 2483 (42.64%) | 15,313 (30.37%) | |
| Other | 2463 (5.52%) | 330 (5.67%) | 2793 (5.54%) | |
| Admission type, n (column %) | <0.0001 | |||
| Elective | 10,017 (22.46%) | 1592 (27.34%) | 11,609 (23.02%) | |
| Labor and delivery | 14,120 (31.66%) | 1057 (18.15%) | 15,177 (30.1%) | |
| Transfer | 3427 (7.68%) | 560 (9.62%) | 3987 (7.91%) | |
| Urgent | 17,037 (38.2%) | 2614 (44.89%) | 19,651 (38.97%) | |
| Category total | 44,601 (100%) | 5823 (100%) | 50,424 (100%) |
*Wilcoxon rank-sum for Cont var and Chi-Sq for categorical values
Table 2.
Demographic of LEP Patients with Different Match Strategies at the Encounter Level
| LEP patients | |||||
|---|---|---|---|---|---|
| Successful match to LIS call log | No match to LIS call log | ||||
| Exact match only | Probabilistic match only | Had both exact and probabilistic matches* | Cannot be found on call log | Total LEP n (column %) |
|
| (n) | |||||
| Age(years) Mean(+SD) | 59.93 (22.68) | 32.05 (35.31) | 64.64 (20.82) | 38.96 (33.40) | 53.77 (28.58) |
| Race, n (row %) | |||||
| Asian | 559 (61.29%) | 31 (3.4%) | 176 (19.3%) | 146 (16.01%) | 912 (15.66%) |
| Black | 48 (30.97%) | 19 (12.26%) | 22 (14.19%) | 66 (42.58%) | 155 (2.66%) |
| Hispanic | 2035 (63.08%) | 95 (2.94%) | 509 (15.78%) | 587 (18.2%) | 3226 (55.4%) |
| Other | 275 (39.74%) | 81 (11.71%) | 94 (13.58%) | 242 (34.97%) | 692 (11.88%) |
| White | 217 (25.89%) | 134 (15.99%) | 94 (11.22%) | 393 (46.9%) | 838 (14.39%) |
| Languages, n (row %) | |||||
| Arabic | 48 (50%) | 7 (7.29%) | 19 (19.79%) | 22 (22.92%) | 96 (1.65%) |
| Bengali | 96 (60%) | 1 (0.63%) | 37 (23.13%) | 26 (16.25%) | 160 (2.75%) |
| Chinese | 384 (69.31%) | 11 (1.99%) | 99 (17.87%) | 60 (10.83%) | 554 (9.51%) |
| Other — not specified | 25 (12.44%) | 37 (18.41%) | 14 (6.97%) | 125 (62.19%) | 201 (3.45%) |
| Other — specified | 275 (46.37%) | 49 (8.26%) | 108 (18.21%) | 161 (27.15%) | 593 (10.18%) |
| Other — unspecified | 1 (0.18%) | 151 (27.06%) | 1 (0.18%) | 405 (72.58%) | 558 (9.58%) |
| Russian | 132 (54.55%) | 12 (4.96%) | 58 (23.97%) | 40 (16.53%) | 242 (4.16%) |
| Sign language | 16 (53.33%) | 0 (0%) | 8 (26.67%) | 6 (20%) | 30 (0.52%) |
| Spanish | 2157 (63.65%) | 92 (2.71%) | 551 (16.26%) | 589 (17.38%) | 3389 (58.2%) |
| Sex, n (row %) | |||||
| Female | 1658 (55.41%) | 176 (5.88%) | 422 (14.1%) | 736 (24.6%) | 2992 (51.38%) |
| Male | 1476 (52.14%) | 184 (6.5%) | 473 (16.71%) | 698 (24.66%) | 2831 (48.62%) |
| Insurance, n (row%) | |||||
| Commercial | 391 (35.55%) | 127 (11.55%) | 101 (9.18%) | 481 (43.73%) | 1100 (18.89%) |
| Medicaid | 1174 (61.47%) | 86 (4.5%) | 264 (13.82%) | 386 (20.21%) | 1910 (32.8%) |
| Medicare | 1468 (59.12%) | 102 (4.11%) | 474 (19.09%) | 439 (17.68%) | 2483 (42.64%) |
| Other | 101 (30.61%) | 45 (13.64%) | 56 (16.97%) | 128 (38.79%) | 330 (5.67%) |
| Admission type, n (row%) | |||||
| Elective | 1010 (63.44%) | 48 (3.02%) | 244 (15.33%) | 290 (18.22%) | 1592 (27.34%) |
| Labor and delivery | 220 (20.81%) | 189 (17.88%) | 45 (4.26%) | 603 (57.05%) | 1057 (18.15%) |
| Transfer | 311 (55.54%) | 27 (4.82%) | 131 (23.39%) | 91 (16.25%) | 560 (9.62%) |
| Urgent | 1593 (60.94%) | 96 (3.67%) | 475 (18.17%) | 450 (17.21%) | 2614 (44.89%) |
| Category total | 3134 (53.82%) | 360 (6.18%) | 895 (15.37%) | 1434 (24.63%) | 5823 (100%) |
Next, we examined the number of LIS calls within each of the top five languages using the exact match method. Based on patients’ preferred language, Spanish had the highest count with 2737 calls, followed by Chinese with 493 calls. In contrast, Arabic had a notably lower count, recording only 58 calls (Appendix Table 1).
Using the probabilistic match approach, the Levenshtein distance scores varied between 71 and 93%, with a median of 86.0%, mean of 84.9%, and a standard deviation of 7.5. This method yielded results that mirrored the exact match findings: Spanish had the highest number of calls at 845, followed by Chinese with 134 (Table 3). Notably, these figures were lower compared to the exact match counts. The alignment between the preferred language from patient registration data and the LIS call log was robust for exact matches exceeding 94% for all languages, except for “Other - Specified” which was at 89.9% (Appendix Table 2). In probabilistic matches, the concordance between patient preferred language and LIS call log language was lower, ranging from 53.9% for Arabic to 71.6% for Russian (Table 4).
Table 3.
Number of probabilistic matched LIS Calls by Preferred Language Among Patients Whose Preferred Language Was Not English
| Patient’s preferred language | No. of calls from probabilistic match | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 10 to ≤20 | Total | |
| Arabic | 0 | 18 | 6 | 2 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 26 |
| Bengali | 0 | 39 | 3 | 1 | 1 | 1 | 0 | 0 | 1 | 0 | 0 | 0 | 46 |
| Chinese | 0 | 105 | 17 | 3 | 3 | 2 | 2 | 1 | 0 | 0 | 1 | 0 | 134 |
| Other—specified | 0 | 103 | 14 | 5 | 3 | 1 | 1 | 0 | 0 | 0 | 1 | 0 | 128 |
| Russian | 0 | 53 | 11 | 2 | 0 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | 67 |
| Sign language | 0 | 7 | 2 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 9 |
| Spanish | 0 | 710 | 85 | 25 | 12 | 5 | 1 | 3 | 0 | 1 | 2 | 1 | 845 |
| Total | 0 | 1035 | 138 | 38 | 19 | 9 | 5 | 4 | 1 | 1 | 4 | 1 | 1255 |
Table 4.
Concordance of Probabilistic Language Match to Preferred Language Preferred Language Was Not English
| Patient’s preferred language | Probabilistic match languages | |||||||
|---|---|---|---|---|---|---|---|---|
| Arabic | Bengali | Chinese | Other —specified | Russian | Sign language | Spanish | Total | |
| Arabic | 14(53.85%) | 0(0%) | 2(1.49%) | 4(3.13%) | 0(0%) | 0(0%) | 6(0.71%) | 26 |
| Bengali | 0(0%) | 27(58.7%) | 2(1.49%) | 1(0.78%) | 0(0%) | 0(0%) | 8(0.95%) | 38 |
| Chinese | 0(0%) | 0(0%) | 90(67.16%) | 5(3.91%) | 0(0%) | 0(0%) | 15(1.78%) | 110 |
| Other —“unknown” | 2(7.69%) | 8(17.39%) | 11(8.21%) | 17(13.28%) | 8(11.94%) | 0(0%) | 106(12.54%) | 152 |
| Other—“unspecified” | 0(0%) | 1(2.17%) | 6(4.48%) | 8(6.25%) | 1(1.49%) | 0(0%) | 35(4.14%) | 51 |
| Other— specified | 2(7.69%) | 1(2.17%) | 9(6.72%) | 75(58.59%) | 3(4.48%) | 0(0%) | 67(7.93%) | 157 |
| Russian | 1(3.85%) | 0(0%) | 1(0.75%) | 1(0.78%) | 48(71.64%) | 1(11.11%) | 18(2.13%) | 70 |
| Sign language | 0(0%) | 0(0%) | 1(0.75%) | 0(0%) | 0(0%) | 6(66.67%) | 1(0.12%) | 8 |
| Spanish | 7(26.92%) | 9(19.57%) | 12(8.96%) | 17(13.28%) | 7(10.45%) | 2(22.22%) | 589(69.7%) | 643 |
| Total | 26 | 46 | 134 | 128 | 67 | 9 | 845 | 1255 |
In essence, the probabilistic method matched an additional 1255 LIS calls to patient encounters (Table 4). When matching was based on the alignment of preferred language with the LIS call log language, 849 more LIS calls were added to patient encounters (Table 4). The overall mean call duration was recorded at 11.52 min per call, per encounter (Appendix Table 3). There was a noteworthy correlation between the frequency of a language spoken in our hospital to the number of LIS calls: languages spoken more frequently like Chinese had 2.42 calls/day, while the “Other - Unspecified” category had a lower rate at 0.57 calls/day (Appendix Table 4).
DISCUSSION
Our analysis reveals that 83.10% (4389/5823) of matches were made either through exact (3134/5823, OR 53.8%) or probabilistic (1434/5823, OR 24.6%) methods, utilizing MRN and date of visit. Despite the inherent risk of errors in probabilistic matching, the inclusion of 1255 more data points can bolster the sample size, reduce type II error (β), and subsequently, amplify statistical power (1-β).20,23,24
Historically, LEP research has indicated that deploying language-concordant care models enhances patient outcomes and trims healthcare expenses for LEP patients.25,26 However, few have provided scalable methods for monitoring compliance with LIS usage. Our methodology bridges this gap, matching patient encounters within a vast health system with language interpreter service call logs. This association not only paints a fuller image of the interpreter services utilized but also serves as a benchmark for assessing care quality.
Our study provides insights about language service preferences, frequency, and duration. Furthermore, it suggests whether the apparent underutilization of interpreters stems from a genuine lack of usage of LIS or merely from the failure to document LIS use by healthcare providers. Additionally, by examining the utilization of specific language services across different hospital locations and comparing it with overarching hospital system data, we gain a clearer understanding of the patterns of LIS usage. Rather than simply determining if patients speak English, our research delves deeper, highlighting specific languages that may face challenges in obtaining interpreting services. In our study, Arabic had the lowest match to the medical records among the commonly spoken languages, while Chinese had the best match.
Difficulties in monitoring actual LIS usage have hindered the development of standards regarding the frequency of interpreter service use. As evidenced by our data, there is an average of about one call per day for most language groups in the inpatient setting. Considering the number of providers, nurses, technicians, and other staff members who interact with an inpatient during a single day of hospitalization, one call a day is insufficient. LIS plays a pivotal role at key junctures during a patient’s stay—such as during initial consultations, procedures, and discharge instructions. Establishing a minimum standard provides a framework against which healthcare systems can evaluate their performance and enhance their interpreter utilization The data is still emerging as to what might be an appropriate base standard for number of calls per day, but as evidenced in this study there is currently either a lack of interpreting services or documentation of interpreting services.
While an ideal measure of compliance would be the exact match, our probabilistic match algorithm estimates match with more than 50% concordance for all languages. From our results, Russian and Spanish had the highest concordance with preferred language and the language recorded in the call log. This might indicate that our model works better for these specific languages, which could be due to prevalent documentation and possibly robust LIS utilization. On the flip side, languages like Bengali and Chinese might be more prone to the inadequacies of interpreter services, exposing these patients to potential communication challenges. Such disparities underscore the importance of tailoring approaches based on the nuanced requirements of diverse linguistic communities. The study further highlights disparities in call durations across different languages, raising pivotal questions about language equity for specific subgroups. The difficulty in quantifying the frequency of language service utilization and of deciphering what languages are being used may have led to past research focusing on language research as a dichotomous variable with LEP patients and non-LEP patients. This underlines the importance of moving away from this binary and exploring the needs of specific language communities.
It is important to recognize that minority communities often harbor historical skepticism towards medical institutions.26,27 Inadequate provision of culturally and linguistic appropriate language services can exacerbate this mistrust, which has been linked with suboptimal health outcomes in past studies.28 Our insights underline the disparities in services rendered to LEP patients both overall and within specific linguistic factions. This methodology also provides a framework to base health outcomes LEP research which can be critical in addressing this skepticism towards medical institutions. The Department of Health and Human Services (HHS) outlines culturally and linguistically appropriate services (CLAS) standards for health organizations to “offer language assistance to individuals who have limited English proficiency and/or other communication needs, at no cost to them, to facilitate timely access to all health care and services.” 14 As healthcare organizations struggle to comply with the CLAS standards, it remains to be determined where the onus of funding these essential LIS calls should fall. When LEP patients require LIS during clinical visits, the duration is typically elongated compared to encounters with English-speaking patients. Despite this additional commitment, there is no commensurate financial compensation for the additional time spent. While the federal government mandates LIS for LEP patients, most commercial payers do not reimburse for LIS and, presently, only a handful of states provide reimbursement for these services via their Medicaid or CHIP programs and even among those that provide reimbursement, the compensation often does not meet the actual expenses.29 With a rising LEP demographic, sustainable financial solutions are paramount to ensure that LEP patients receive the appropriate communication that high-quality care requires.
While our model begins to establish quantitative measures of interpreter use, there are limitations, and we categorize these into five major categories. First, a patient may have their preferred language documented as English in their chart yet still utilize interpreter services, rendering the MRN unmatchable with a call log. This may be due to different levels of fluency or health literacy. Since our method uses data from patients with preferred language other than English in their charts, these interpreter interactions would not be included in our data. Second, conversely, a patient with a preferred language other than English in their chart may speak some English and thus choose to forgo interpreter services. Again, fluency plays a large role in how comfortable a patient may be with or without an interpreter and further emphasizes the need to assure individualized care. Hospitals risk overestimating their interpreter requirements in such scenarios. Third, certified language concordant providers, especially those proficient in widely spoken languages like Spanish and Chinese, might not use interpreters. It should be noted that the use of uncertified bilingual providers as interpreters is a violation of federal mandates.12,30 This might not be the case for languages less prevalent in the community. While it is important to have a language certification assessment to make sure clinicians have the proper level of fluency to practice in a certain language, there is currently no agreed-upon method of doing this. There is no current standardization for certifying bilingual providers, which is important when evaluating the limitations of this analysis. Fourth, some LEP patients might decline formal interpreter services, preferring family, or friends—a practice more prevalent in specific cultural contexts (Table 2). Lastly, a significant proportion of our LIS call logs did not contain an MRN, which can arise from various factors. For instance, emergency calls might not capture a patient’s MRN. Similarly, using LIS to respond to inquiries from LEP visitors at the front desk may not include an MRN. Of the 183,655 call log entries recorded in 2020, only 21,731 were successfully matched to hospital visits. The remaining entries contained medical record numbers (MRNs) that did not correspond with those in the main hospital visit records. Since this only captured 12% of call log entries, this is a key limitation of the study. The underlying reasons for these mismatches remain unclear, warranting further investigation. In conclusion, our study demonstrated the feasibility of monitoring LIS usage through matching LIS calls to patient encounters. Despite its limitations, our method allows healthcare organizations to introduce LIS usage as part of their quality assurance programs.
Patients with limited English proficiency (LEP) often encounter barriers when accessing healthcare, leading to suboptimal health outcomes and an increased risk of medical errors.4,5,9,10,12,31–36 A significant obstacle in this regard is the challenge of communication, which is further exacerbated by inadequate or ineffective language interpreter services. To improve these services, it is crucial to have a comprehensive understanding of the current system and ensure compliance with interpreter service utilization. However, achieving this goal is challenging without standardized documentation of interpreter use.
Our matching methodology provides a framework for collating data on language interpretation services, paving the way for enhancing care for patients with limited English proficiency. As the field progresses, it is imperative to move beyond viewing English proficiency in binary terms. The complexities of different languages spoken, varying degrees of fluency, and diverse health literacy levels must be considered to ensure the highest standard of care for all patients. By recognizing and addressing these complexities, healthcare systems can better meet the needs of linguistically diverse patient populations and improve health outcomes.
Supplementary Information
Below is the link to the electronic supplementary material.
Declarations:
Conflict of Interest:
The authors declare that they do not have a conflict of interest.
Footnotes
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
- 1.Shin H, Kominski R. Language Use in the United States: 2007 http://www.census.gov/prod/2010pubs/acs-12.pdf2010 [cited 2011 November 12]. Available from: http://www.census.gov/prod/2010pubs/acs-12.pdf.
- 2.Proctor K, Wilson-Frederick SM, Haffer SC. The Limited English Proficient Population: Describing Medicare, Medicaid, and Dual Beneficiaries. Health Equity. 2018;2(1):82-9. 10.1089/heq.2017.0036. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Ramirez N, Shi K, Yabroff KR, Han X, Fedewa SA, Nogueira LM. Access to Care Among Adults with Limited English Proficiency. J Gen Intern Med. 2023;38(3):592-9. 10.1007/s11606-022-07690-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Yeheskel A, Rawal S. Exploring the 'Patient Experience' of Individuals with Limited English Proficiency: A Scoping Review. J Immigr Minor Health. 2019;21(4):853-78. 10.1007/s10903-018-0816-4. [DOI] [PubMed] [Google Scholar]
- 5.Diamond LC, Schenker Y, Curry L, Bradley EH, Fernandez A. Getting by: underuse of interpreters by resident physicians. J Gen Intern Med. 2009;24(2):256-62. 10.1007/s11606-008-0875-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Karliner LS, Jacobs EA, Chen AH, Mutha S. Do professional interpreters improve clinical care for patients with limited English proficiency? A systematic review of the literature. Health Serv Res. 2007;42(2):727-54. 10.1111/j.1475-6773.2006.00629.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Diamond Lc Fau - Tuot DS, Tuot Ds Fau - Karliner LS, Karliner LS. The use of Spanish language skills by physicians and nurses: policy implications for teaching and testing. J Gen Intern Med. 2012;27(1525-1497 (Electronic)). doi: D - NLM: PMC3250531 [Available on 01/01/13] EDAT- 2011/07/21 06:00 MHDA- 2012/09/21 06:00 CRDT- 2011/07/21 06:00 PMCR- 2013/01/01 00:00 PHST- 2011/01/30 [received] PHST- 2011/06/13 [accepted] PHST- 2011/05/26 [revised] PHST- 2011/07/20 [aheadofprint] AID - 10.1007/s11606-011-1779-5 PST - ppublish. [DOI] [PMC free article] [PubMed]
- 8.Divi C, Koss RG, Schmaltz SP, Loeb JM. Language proficiency and adverse events in US hospitals: a pilot study. Int J Qual Health Care. 2007;19(2):60-7. 10.1093/intqhc/mzl069. [DOI] [PubMed] [Google Scholar]
- 9.Kim EJ, Kim T, Paasche-Orlow MK, Rose AJ, Hanchate AD. Disparities in Hypertension Associated with Limited English Proficiency. J Gen Intern Med. 2017;32(6):632-9. 10.1007/s11606-017-3999-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Ngai KM, Grudzen CR, Lee R, Tong VY, Richardson LD, Fernandez A. The Association Between Limited English Proficiency and Unplanned Emergency Department Revisit Within 72 Hours. Ann Emerg Med. 2016;68(2):213-21. 10.1016/j.annemergmed.2016.02.042. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Guidance to Federal Financial Assistance Recipients Regarding Title VI and the Prohibition Against National Origin Discrimination Affecting Limited English Proficient Persons. United States Depatment of Health and Human Services.
- 12.Chen AH, Youdelman MK, Brooks J. The legal framework for language access in healthcare settings: Title VI and beyond. J Gen Intern Med. 2007;22 Suppl 2:362-7. 10.1007/s11606-007-0366-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Ebert DA, Heckerling PS. Communication with deaf patients. Knowledge, beliefs, and practices of physicians. JAMA (Chicago, Ill). 1995;273(3):227-9. [PubMed]
- 14.Grubbs V, Chen AH, Bindman AB, Vittinghoff E, Fernandez A. Effect of awareness of language law on language access in the health care setting. J Gen Intern Med. 2006;21(7):683-8. 10.1111/j.1525-1497.2006.00492.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Nailon RE. The assessment and documentation of language and communication needs in healthcare systems: current practices and future directions for coordinating safe, patient-centered care. Nurs Outlook. 2007;55(6):311-7. 10.1016/j.outlook.2007.04.005. [DOI] [PubMed] [Google Scholar]
- 16.The Joint Commission: Advancing Effective Communication, Cultural Competence, and Patient- and Family-Centered Care: A Roadmap for Hospitals.2010.
- 17.The Joint Commission: R3 Report Requirement, Rationale, Reference2011(1).
- 18.Ramos R, Davis JL, Ross T, Grant CG, Green BL. Measuring Health Disparities and Health Inequities: Do You Have REGAL Data? Quality Management in Healthcare. 2012;21(3):176-87. 10.1097/QMH.0b013e31825e8889. [DOI] [PubMed] [Google Scholar]
- 19.Cohen WW, Ravikumar P, Fienberg SE, editors. A Comparison of String Distance Metrics for Name-Matching Tasks. IIWeb; 2003.
- 20.Cohen J. Statistical Power Analysis for the Behavioral Sciences Lawrence Erlbaum Associates: Hillsdale, NJ; 1988.
- 21.SeatGeek. FuzzyWuzzy: Fuzzy string matching in Python. GitHub repository [cited 2023 Sept 11]; TheFuzz version 0.19.0 [Available from: https://github.com/seatgeek/fuzzywuzzy.
- 22.Levenshtein VI. Binary codes capable of correcting deletions, insertions, and reversals. Soviet Physics Doklady. 1966;10(8):707-10. [Google Scholar]
- 23.Sayers A, Ben-Shlomo Y, Blom AW, Steele F. Probabilistic record linkage. Int J Epidemiol. 2016;45(3):954-64. 10.1093/ije/dyv322. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Zech J, Husk G, Moore T, Shapiro JS. Measuring the Degree of Unmatched Patient Records in a Health Information Exchange Using Exact Matching. Appl Clin Inform. 2016;7(2):330-40. 10.4338/aci-2015-11-ra-0158. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Nápoles AM, Santoyo-Olsson J, Karliner LS, Gregorich SE, Pérez-Stable EJ. Inaccurate Language Interpretation and Its Clinical Significance in the Medical Encounters of Spanish-speaking Latinos. Med Care. 2015;53(11):940-7. 10.1097/mlr.0000000000000422. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Elkind MS, Carty CL, O'Meara ES, Lumley T, Lefkowitz D, Kronmal RA, Longstreth WT, Jr. Hospitalization for infection and risk of acute ischemic stroke: the Cardiovascular Health Study. Stroke. 2011;42(7):1851-6. 10.1161/strokeaha.110.608588. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Bazargan M, Cobb S, Assari S. Discrimination and Medical Mistrust in a Racially and Ethnically Diverse Sample of California Adults. Ann Fam Med. 2021;19(1):4-15. 10.1370/afm.2632. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Benkert R, Peters RM, Clark R, Keves-Foster K. Effects of perceived racism, cultural mistrust and trust in providers on satisfaction with care. J Natl Med Assoc. 2006;98(9):1532-40. [PMC free article] [PubMed] [Google Scholar]
- 29.Medicaid.gov. Translation and Interpretation Services [cited 2023 Sept 11]. Available from: https://www.medicaid.gov/medicaid/financial-management/medicaid-administrative-claiming/translation-and-interpretation-services/index.html.
- 30.People H, Prevention OoD, Promotion H. Washington, DC: US Department of Health and Human Services. Public Health Service. 2000. [Google Scholar]
- 31.Fernandez A. Improving the quality of informed consent: it is not all about the risks. Annals Int Med. 2010;153(5):342-3. 10.1059/0003-4819-153-5-201009070-00011 [DOI] [PubMed] [Google Scholar]
- 32.Ginde AA, Clark S, Camargo CA, Jr. Language barriers among patients in Boston emergency departments: use of medical interpreters after passage of interpreter legislation. J Immigr Minor Health. 2009;11(6):527-30. 10.1007/s10903-008-9188-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Ku L, Flores G. Pay now or pay later: providing interpreter services in health care. Health affairs. 2005;24(2):435-44. [DOI] [PubMed] [Google Scholar]
- 34.Baker DW, Parker RM, Williams MV, Coates WC, Pitkin K. Use and effectiveness of interpreters in an emergency department. JAMA J Am Med Assoc 1996;275(10):783-8. [PubMed]
- 35.Bernstein J, Bernstein E, Dave A, Hardt E, James T, Linden J, Mitchell P, Oishi T, Safi C. Trained medical interpreters in the emergency department: effects on services, subsequent charges, and follow-up. J Immigrant Health. 2002;4(4):171-6. 10.1023/a:1020125425820. [DOI] [PubMed] [Google Scholar]
- 36.Flores G, Abreu M, Barone CP, Bachur R, Lin H. Errors of medical interpretation and their potential clinical consequences: a comparison of professional versus ad hoc versus no interpreters. Ann Emerg Med. 2012;60(5):545-53. 10.1016/j.annemergmed.2012.01.025. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.