Limited English Proficiency and Disparities in Health Care Engagement Among Patients With Breast Cancer

PURPOSE:

Race and ethnicity have been shown to affect quality of cancer care, and patients with low English proficiency (LEP) have increased risk for serious adverse events. We sought to assess the impact of primary language on health care engagement as indicated by clinical trial screening and engagement, use of genetic counseling, and communication via an electronic patient portal.

METHODS:

Clinical and demographic data on patients with breast cancer diagnosed and treated from 2013 to 2018 within the Stanford University Health Care system were compiled via linkage of electronic health records, an internal clinical trial database, and the California Cancer Registry. Logistic and linear regression models were used to evaluate for association of clinical trial engagement and patient portal message rates with primary language group.

RESULTS:

Patients with LEP had significantly lower rates of clinical trial engagement compared with their English-speaking counterparts (adjusted odds ratio [OR], 0.29; 95% CI, 0.16 to 0.51). Use of genetic counseling was similar between language groups. Rates of patient portal messaging did not differ between English-speaking and LEP groups on multivariable analysis; however, patients with LEP were less likely to have a portal account (adjusted OR, 0.89; 95% CI, 0.83 to 0.96). Among LEP subgroups, Spanish speakers were significantly less likely to engage with the patient portal compared with English speakers (estimated difference in monthly rate: OR, 0.43; 95% CI, 0.24 to 0.77).

CONCLUSION:

We found that patients with LEP had lower rates of clinical trial engagement and odds of electronic patient portal enrollment. Interventions designed to overcome language and cultural barriers are essential to optimize the experience of patients with LEP.

INTRODUCTION

There has been increasing interest in quality and access to cancer care, along with increasing awareness of disparities that contribute to unequal outcomes.¹ Race and ethnicity have been shown to affect access to and quality of care.² Language is crucial in the practice of medicine and especially in oncology, where clinicians rely on verbal exchanges to provide information and engage patients in shared decision making.^3,4 Approximately 25 million people in the United States have limited English proficiency (LEP), posing additional challenges to communication among the patient, their caregivers, and clinicians.^5,6 In a national study based on US census data, primary language (non–English-speaking) most strongly predicted an unfavorable rating of physician communication. Studies have also shown that language-concordant care improves outcomes.⁶

Physician-patient communication is essential for cancer care, for educating patients about their disease and managing their symptoms, discussing prognosis, and for assisting in navigation of the complex health care environment. Communication is also key in establishing a trusting relationship that facilitates dialogue about participation in clinical trials. Prior research has associated LEP status with a two-fold increase in risk for serious medical events compared with English-speaking patients^7,8 and may contribute to misunderstandings.⁹ Communication in oncology practice also extends beyond the office setting and includes electronic health care portals. Portals allow bidirectional messaging and are commonly used for scheduling visits and refilling medications; they also provide patients with a channel to message their clinicians expressing concern or seeking guidance.¹⁰ However, such portals usually assume English proficiency, a requirement that may unintentionally exacerbate disparities in access to care.¹¹

We sought to assess the impact of primary language on health care engagement as indicated by three areas of cancer care that rely on communication: (1) screening and/or enrolling patients in cancer clinical trials, (2) access to genetic counseling, and (3) communication using an established electronic patient portal, in a patient sample treated for breast cancer at a tertiary academic cancer center (Stanford Cancer Center). We hypothesized that patients with LEP would have a lower recorded number of screening events for participation in clinical research, fewer genetic counseling visits, and a lower volume of clinically related messages captured in the electronic portal.

METHODS

Study Population

We queried the Oncoshare database, which combines data from Stanford Health Care’s electronic health records (EHR) (including outpatient and inpatient visits, physicians’ text notes, imaging reports, infused drug records, records of telephone notes, and portal use and content) and the California Cancer Registry (including clinical variables such as stage at diagnosis and first course of treatment, and sociodemographic variables such as insurance and neighborhood-level socioeconomic status [SES]).^12,13 We included patients with breast cancer diagnosed from January 1, 2013, to December 31, 2018, who were treated at Stanford Health Care. Patients with LEP were defined as patients with either the primary language in the demographics section in the EHR listed as a language other than English and/or a listed yes for the needs interpreter field. Baseline characteristics, clinical trial outcomes, and genetic counseling outcomes were reported overall and by primary language group.

Data Sources

We linked the Oncoshare database with the Stanford Cancer Center’s clinical trial database (OnCore Enterprise Research, a comprehensive clinical research management system, Advarra Inc, Columbia, MD). We defined clinical trial engagement as any contact with the clinical trial team, including if patients were screened and found eligible or not for a trial, and/or enrolled in a trial. The OnCore database is the comprehensive clinical trial repository for Stanford cancer patients, and thus patients without any data on linkage were assumed to not have formal clinical trial engagement. All clinical trials (observational and interventional) were considered. We also conducted a retrospective chart review of the subset of patients with LEP who met our criteria of clinical trial engagement to assess the documentation of professional interpreter use during conversations about clinical trials.

The Stanford Clinical Cancer Genetics and Genomics Program maintains a separate research database of all patients who have had at least one consultation with a genetic counselor, most of whom undergo genetic testing. The Stanford Cancer Genetics Research Database also links to the Oncoshare database.

We measured electronic portal usage (MyHealth) by number of messages exchanged in the EHR from patient to the clinical team and vice versa and classified as previously described.¹⁴ Patient portal activation occurs in two steps: (1) a link is provided by the reception staff and (2) the patient then activates the account via this link on their own electronic device. Message types included those deemed of clinical relevance: general medical questionnaire, patient medical advice request, patient appointment change request, and patient medication renewal request. We excluded any automated messages.

Statistical Analysis

To evaluate whether LEP was associated with lower rates of clinical trial engagement, a logistic regression model was fit to clinical trial enrollment (yes or no) as a function of primary language group (English v LEP). This model was refit adjusting for age at diagnosis, race, SES, tumor stage, any radiation therapy use, any chemotherapy use, and insurance type. Odds ratios (ORs) and 95% CIs were reported for both the unadjusted and adjusted models.

A chi-squared test was used to determine whether there was an association between patients receiving genetic counseling and primary language group.

A logistic regression model was fit to assess whether a MyHealth account was created as a function of primary language group and to determine whether the odds of account creation differed by language group. This model was refit adjusting for the covariates described above.

To compare messages between patients with varying MyHealth active times, a monthly MyHealth message rate was calculated by summing the number of MyHealth messages over the number of months that MyHealth was active. Each patient’s MyHealth active time was defined as the number of months they had an active MyHealth account between breast cancer diagnosis and last follow-up. Only patients with at least 3 months of active MyHealth time were included to observe a sufficient time interval for meaningful communication. To determine whether the overall MyHealth message rate differed between English speakers and patients with LEP, a linear regression model was fit to log-transformed MyHealth message rate as a function of language spoken. Associations between MyHealth messaging rates and age at diagnosis were assessed using Pearson’s correlation coefficient.

We were also interested in evaluating whether the rate of MyHealth messages differed over time by language group. A mixed-effect linear regression model was fit to log-transformed number of messages per year as a function of primary language group, number of years post-breast cancer diagnosis, and log-transformed number of months of MyHealth active time within each year, with a random effect for patient. This model was refit adjusting for the same covariates as described above. Change in the number of MyHealth messages per year and 95% CI were reported for all MyHealth models.

Primary language group was further investigated in the clinical trial and MyHealth analyses comparing English-speaking versus LEP subgroups. These models were refit breaking down the LEP group into Asian, Spanish, and Other language subgroups.

All data were deidentified before analysis. A P value of < .05 was considered statistically significant. Statistical analysis was performed using R version 4.0.2 (R Foundation for Statistical Computing, Vienna, Austria). This work was approved by the Institutional Review Board of Stanford University.

RESULTS

Baseline Characteristics

Of the 1,849 patients included in this study, 196 (11%) were LEP (Table 1). The median age at breast cancer diagnosis was 52 years, with most patients being non-Hispanic White (57%). More than half (58%) of the patients had private insurance, most were stage I or II (61%), and more than half received chemotherapy or radiation therapy. Although most baseline characteristics were similar between primary language groups, there were notable differences in the racial and ethnic makeup of the cohorts, as expected, as well as a higher percentage of English speakers with private insurance compared with non-English speakers (61% v 29%, respectively). Race, ethnicity, and insurance demographics from the Stanford Cancer Center have been previously described and follow a similar distribution to our findings.^6,12,14

TABLE 1.

Baseline Demographics

Clinical Trial and Genetic Counseling Engagement

Patients with LEP had lower rates of clinical trial engagement (including assessing for eligibility for clinical trial, even if the patient was not enrolled) compared with English-speaking patients (11.2% v 22.7%). In the adjusted model, the LEP group had 0.3 times the odds of having clinical trial engagement as the English-speaking group (95% CI, 0.16 to 0.51, P < .001). These results remained statistically significant after adjusting for age at diagnosis, race, SES, tumor stage, radiation therapy use, chemotherapy use, and insurance type. Subgrouping the patients with LEP into Spanish, Asian, and Other, we found that all three LEP subgroups were significantly less likely to have clinical trial engagement in the adjusted model (Data Supplement, online only). In a retrospective chart review of the 22 patients with LEP who met study criteria for clinical trial engagement, 18 patients were noted to have been approached regarding a clinical trial during a clinic visit (the other four patients were screened for trial but a discussion was not documented). Of the 18 patients, seven (38.9%) had documented use of a professional interpreter for the initial clinical trial discussion (data not shown). For seven patients who provided consent for clinical trial enrollment, five had a professional interpreter present during the consent process, whereas the other two patients did not have a documented interpreter present. Both of these patients were offered an interpreter but opted to have their family member interpret during the informed consent process, per clinical notes.

The percentages of patients receiving genetic counseling were similar between the LEP and English-speaking groups (35.2% v 40.7%, Pearson χ² P = .14).

MyHealth Message Engagement

Of the cohort of 1,849 patients, 1,482 (80%) had a MyHealth account created during the observation period. Seventy percentage of patients with LEP had a MyHealth account compared with 81% of English-speaking patients, corresponding to an OR of 0.89 (95% CI, 0.84 to 0.95, P < .001). Patients with LEP had significantly lower odds of having a MyHealth account compared with English speakers after adjusting for the aforementioned covariates of interest. Lower odds of MyHealth accounts were seen among Spanish (adjusted OR, 0.85; 95% CI, 0.74 to 0.98) and Other (adjusted OR, 0.80; 95% CI, 0.70 to 0.89) LEP subgroups compared with English speakers (Data Supplement).

Of the original 1,849 patients, we excluded patients who did not have an activated MyHealth account (n = 367 excluded). We further excluded anyone whose MyHealth account activation ended before breast cancer diagnosis or started after last follow-up (n = 78 excluded) and additionally excluded patients with < 3 months of active time (n = 111 excluded). Therefore, the analytic cohort for this subset was n = 1,293. Baseline characteristics of this subset were similar to those of the full cohort.

Although median MyHealth active time was similar between the LEP and English-speaking groups (40.9 months v 36.9 months, respectively), the median number of total messages was higher for English-speaking (32) versus patients with LEP (14). The median monthly message rate for English speakers was 0.87 messages per month compared with 0.49 messages per month for patients with LEP (Fig 1).

FIG 1.

Distribution of MyHealth message rate by language group. Boxplots of the monthly MyHealth message rate by primary language group. Line in the center of the box represents the median. Each dot represents a patient.

In the unadjusted model, the average MyHealth monthly message rate for non-English patients was 37% lower than the rate for English patients (estimated difference in message rate, 0.63; 95% CI, 0.50 to 0.80; P < .001; Data Supplement). However, after adjusting for potential confounders, this difference narrowed and was no longer significant (estimate, 0.80; 95% CI, 0.60 to 1.05; P = .12). Looking among LEP subgroups, Spanish speakers had the lowest monthly message rates, followed by Asian languages and Other languages; Spanish speakers were the only subgroup to have significantly lower rates of messages compared with English speakers, with an adjusted OR of 0.43 (95% CI, 0.24 to 0.77) (Data Supplement).

We assessed the impact of age at diagnosis to activation and usage of MyHealth. We did not see an effect of age on whether or not a MyHealth account was activated. There was a statistically significant yet weak negative correlation between the messaging rate and age at diagnosis (Pearson’s χ², 0.12; 95% CI, −0.17 to −0.06; P < .001). In our adjusted models, we found that this association persists in the presence of covariates for MyHealth messaging rate (estimate, 0.989; 95% CI, 0.982 to 0.996).

Figure 2 summarizes the number of MyHealth messages up to 8 years post-breast cancer diagnosis, by language group. Rates of monthly MyHealth messages decreased over time in both language groups. In both the unadjusted and adjusted models, patients with LEP had a lower number of monthly MyHealth messages over time compared with English-speaking patients adjusting for active time and years post-diagnosis, but was only significantly different in the unadjusted model (Data Supplement). Adjusting for covariates of interest, patients with LEP had a 20% lower rate of messages compared with English-speaking patients over the entire observation period; however, this finding was not statistically significant (estimated difference in number of messages over time, 0.80; 95% CI, 0.62 to 1.03). The number of messages decreased by 9% per year post-diagnosis (0.91; 95% CI, 0.90 to 0.93) for all patients. The number of patients contributing to the data per year and the total number of messages per year post-breast cancer diagnosis are shown in the Data Supplement. In evaluating these changes among LEP subgroups, the difference in the rates of messages over time messages appeared to be driven by the Spanish subgroup (Data Supplement).

FIG 2.

MyHealth message rate over time by primary language. Line plots of MyHealth message rates per month over time, colored by primary language group. Lightly colored lines are individual patient trajectories, whereas bolded lines are the aggregated mean message trajectory with respective confidence bands. Y-axis truncated to 10 for better visualization.

DISCUSSION

We found that patients with LEP were approximately 70% less likely to have any recorded indication of clinical trial engagement, even after adjusting for potential confounding factors such as race, SES, and insurance type. We leveraged three existing institutional research databases (breast cancer, clinical trials, and cancer genetics) to minimize missing data, which is an inherent limitation in retrospective studies of clinical encounter data. To our knowledge, this is one of the first studies to explore a range of communication-centered cancer care events by spoken language and, as such, it offers an important window into potential gaps in the care of patients with LEP.

Our work adds to prior studies that have found less clinical trial enrollment of patients with LEP. For example, in a study of more than 19,000 patients treated at a culturally diverse Australian hospital, clinical trial participation was significantly lower in patients with LEP (5.7% v 8.4%; OR, 0.80; P = .001).¹⁵ In a 2018 study based in Los Angeles, CA, of 12,538 index cancer cases identified, 10% preferred a language other than English. Although trial enrollment was similar with English- and Spanish-speaking patients, the study found lower participation for Russian- and Arabic-speaking patients.¹⁶ Our analysis found significantly lower odds of clinical trial engagement for patients with LEP compared with English-speaking patients, even after adjusting for possible confounders. We note that there are many additional factors aside from language that affect clinical trial enrollment, including but not limited to mistrust of research, lack of culturally relevant education about trials, rigid eligibility characteristics, and data collection burden and costs.^17-19

We found documentation of involvement of a professional interpreter in approximately 62% of visits that led to obtaining consent for enrollment and fewer for preparatory visits leading up to recruitment. We found that the remaining patients provided consent without a professional interpreter but were accompanied by a family member, and it is possible that they declined the offer to have professional language interpretation. Although this analysis is limited by a small sample size, the trend shows inconsistent documentation of involvement of professional interpreters in the multiple steps involved in screening potential cancer clinical trial participants.

By contrast and contrary to our hypothesis, genetic counseling rates did not differ by language spoken. Although we do not know the exact reason for this finding, we do note some features of the Stanford Clinical Cancer Genetics program that are common to many such services. For example, licensed genetic counselors, who are the primary patient-facing clinicians in this program, all hold a master’s degree that required extensive coursework in patient communication and counseling; additionally, new patient visits are uniformly scheduled for 1-hour time intervals, which offers sufficient time for translation of complex information. In comparison, discussion and screening for clinical trials are typically embedded within the time allotted for an oncology outpatient visit, during which many other topics need to be covered for clinical decision making. Furthermore, for the past decade, this clinical service has maintained at least one full-time genetic counselor who is a native Spanish speaker and certified in medical interpretation

Last, we found that non–English-speaking patients were less likely to have communication occur through the electronic patient portal messaging, with fewer patients having an active personal electronic messaging portal (MyHealth). In our adjusted model, we did not find a significant difference in rates of monthly messaging; however, differences were significant between Spanish-speaking and English-speaking patients. As expected, messaging was highest in the year post-diagnosis and decreased over time. We note that the current portal does not offer non-English language functionalities, which likely posed a barrier to many patients with LEP.

Our study has some limitations. Extracting clinical data from the electronic medical record for the secondary purpose of research poses some challenges. We aimed to reduce these challenges by using three institutional research databases.²⁰ Since all of these resources include data entry, processing, and curation by dedicated research staff, the quality of data for research purposes is high. We note that our definition of LEP is likely limited by our reliance of EHR classification, as the LEP.gov website estimates the LEP prevalence of 18%-22% in the Stanford catchment area, whereas our cohort had a prevalence of 10.6%.²¹ We likely missed patients who have lower English proficiency but are listed as speaking English, and it is possible that patients with LEP who live in our catchment area receive their care elsewhere. Additionally, we would have liked to explore differences in recruitment to different types of clinical trials, but we did not have consistent classifications in our database.

The analysis of electronic portal messages required management of less structured data. We attempted to reduce bias and improve data quality by counting only messages of clinical value and only periods when a patient had an active account for the messaging portal. We note that there may be inaccuracies in how messaging account activation and end dates were logged; however, it seems unlikely that any such inaccuracies would be ascertained differently according to the language spoken and thus would be unlikely to affect our conclusions. We also acknowledge that the results from the LEP subgroup analyses were not adjusted for multiple comparisons; thus, any significant results stemming from these models should be confirmed in future studies. Finally, this is a single-center study and thus does not represent all US cancer care settings. However, these limitations are balanced by considerable strengths, notably a very diverse patient population with distinct categories of non-English languages that reflect the demographics of the San Francisco Bay Area. This linguistic diversity enabled us to evaluate results across LEP subgroups, which suggests the relevance of our findings beyond a single racial or ethnic population.

This study’s findings have important implications for understanding and improving disparities in cancer care based on preferred language. Key targets for immediate intervention include the lower clinical trial engagement and electronic messaging enrollment of patients with LEP. EHRs currently are largely only available and used in English, with usually limited availability of content in other languages. With increasing use of telemedicine and electronic portals, health systems must focus on increasing access to adequate language interpreter and translation services to address the need for engagement of patients with LEP. There is already preliminary data showing decreasing proportion of patient visits for patients with a non-English language preference after increasing telemedicine implementation.²² The greater equity of genetic counseling receipt warrants further study to pinpoint the key contributing factors that might be extended to other care settings: likely components include focused clinician training in patient communication, adequate appointment time to translate complex information, inclusion of professionals who speak languages that are highly represented in the communities that seek care at the cancer center, and the availability of professional interpreter services.

AUTHOR CONTRIBUTIONS

Conception and design: Mohana Roy, Natasha Purington, Douglas W. Blayney, Allison W. Kurian, Lidia Schapira

Financial support: Allison W. Kurian

Provision of study materials or patients: Mina Liu, Allison W. Kurian

Collection and assembly of data: Mohana Roy, Natasha Purington, Mina Liu, Allison W. Kurian

Data analysis and interpretation: Mohana Roy, Natasha Purington, Allison W. Kurian, Lidia Schapira

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

Limited English Proficiency and Disparities in Health care Engagement Among Patients With Breast Cancer

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/op/authors/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).