Background

The crossover between technology and healthcare services offers an innovative and efficient way to provide medical care tailored to the needs and preferences of patients, streamlining the process of medical care. As an example, online patient portals, tethered to the electronic medical record (EMR), are a new standard for communication and service in the healthcare industry. These portals allow patients to contact provider teams, schedule visits, view visit summaries and medical record information, check test results, access health education, and manage medication prescriptions and refills from the convenience of their Internet-connected device. After almost two decades since the implementation of portals in early adopter systems (i.e., Kaiser Permanente, Group Health), the results are promising [1,2,3]. Health benefits associated with portal use include glycemic control among diabetics, cholesterol reduction (adherence to statins), and improved blood-pressure management [3].

However, recent evidence demonstrates that patients with communication barriers are less likely to access a variety of digital (Internet-connected) tools like the patient portal—as these are almost exclusively designed and implemented with a “mainstream” English-speaking population in mind. It is well documented that digital divides exist in the general US population by race/ethnicity, income, older age, educational attainment, rural residence, and health literacy (which also encompasses English proficiency) [4,5,6,7,8]. The digital divide reflects the structural barriers to digital access disproportionately affecting these vulnerable groups: (1) access to the Internet; (2) knowledge on how to use the Internet; (3) access to Internet-connected devices such as computers, tablets, and/or phones; and (4) knowledge on how to use Internet-connected devices [4,5,6,7,8]. But, in addition, system-design factors such as ease of use and, for example, availability of digital interfaces in multiple languages, affect a patient’s ability to uptake and effectively engage with technology tools in the healthcare system. While non-white race/ethnicity was the strongest negative predictor of patient portal registration in the well-known Kaiser system, uptake further decreased for Spanish-speaking Latinos and minority older adults who were non-English speaking [7]. These findings suggest that the same subgroups of vulnerable populations who have traditionally struggled to access medical care in the USA and have faced stark health disparities compared to the general population, do not utilize technological advancements that could provide efficient disease management—which include remote medication refills through an Internet patient portal [8,9,10,11].

One group that is notably left behind in this divide are the 22.3 million people in the USA who are limited English proficient (LEP)—defined as a “limited ability to listen, speak, read, and write in English” and/or “self-rated English ability of less than “very well.” [12] Patients with limited English proficiency and chronic disease are at increased risk for poor medication control and non-adherence, and yet do not utilize beneficial healthcare tools, like medication refill access through a portal [12,13,14]. Prior research examined the relationship between English proficiency and use of Internet medication refill systems and found that only 21% of LEP patients with chronic illness used the online service, compared to 53% of English-proficient patients [6]. This discrepancy is not surprising, given that the majority of refill services through an online patient portal are exclusively delivered in English—but also because LEP patients are affected by many of the aforementioned sociodemographic factors that exacerbate the digital divide (lower education, lower income, foreign-born status, lower literacy levels, and decreased access to technology) [6.]

This is problematic, as patients who do use the patient portal for medication refills demonstrate higher rates of medication adherence [2]. And despite the known digital divide, a handful of studies show that racial/ethnic minorities and other underserved populations directly benefit (improvements in adherence) from the refill function through an online patient portal, when they utilize it [7, 15]. However, these studies are nonetheless focused on English-speaking patients. No prior studies, that we are aware of, have examined the association between remote Internet refills through a patient portal and medication adherence, among LEP patients—likely because the majority of patient portals are English only. This still begs the question as to how LEP populations are affected by the current, albeit imperfect, portal services that do exist.

The following study addresses this specific gap with regard to medication adherence through the following aims: (1) to examine the relationship between use of an Internet medication refill system and medication adherence among linguistically diverse patients with chronic conditions and (2) to compare this relationship between LEP and English-proficient (EP) patients.

Methods

Setting

The data for this study was collected from 509 adults with chronic conditions in the Group Health Cooperative (GHC), a nonprofit healthcare system serving approximately 600,000 enrollees in the State of Washington. GHC is an early adopter of the EMR—an integrated patient portal (English-only) is available since 2003.

The study used four sources of data: telephone survey, health plan enrollment, inpatient and outpatient claims data (includes interpreter utilization), and electronic medical records linked to respondents. The Institutional Review Boards at GHC and University of California, Los Angeles, approved this study. The datasets generated/analyzed during the current study are not publicly available due to ongoing analyses but are available from the corresponding author on reasonable request.

Participants

A telephone survey about chronic conditions and medication management was conducted between September 2013 and January 2014 in six languages. Inclusion criteria were as follows:

  1. 1.

    18 years or older

  2. 2.

    English, Spanish, Korean, Cantonese, Mandarin or Vietnamese-speaking (six most common languages at Group Health); for patients whose primary language was not English, an additional inclusion criterion was the use of an interpreter during at least one clinical visit between 2005 and 2012

  3. 3.

    Continuous enrollment in Group Health integrated group practice 6 months prior to the beginning of the survey

  4. 4.

    ICD-9 diagnosis of diabetes mellitus (DM), hypertension (HTN), and/or hyperlipidemia (HL)

  5. 5.

    At least one outpatient clinic visit (not urgent care or emergency) within the health system during the last 6 months

Exclusion criteria were diagnoses of Alzheimer’s disease, dementia, renal failure, pregnancy in prior 12 months or an admission to a hospice, hospital, or SNF in prior 12 months. The survey population was a subset of patients randomly selected from a larger cohort of eligible patients with DM, HTN, or HL—representative of the GHC population. A total of 1490 participants (493 EP and 997 LEP) were invited to complete the survey. LEP patients were oversampled to attain representation by each language group, as reflected in the GHC [6].

Primary Outcome, Medication Adherence

Medication adherence was measured using the well-validated continuous measure of medication gaps method (CMG) [16, 17]. CMG is an objective measure of adherence that uses pharmacy filling of medications to measure gaps in patients’ available supply of medications. The CMG adherence measure was based on at least two pharmacy fills on an existing prescription over a 12-month period preceding the survey (medication possession ratio). The CMG percentage is calculated as the number of days a patient had medication available (based on fills) divided by the number of days the patient should have been on medication. We report the proportion of days covered. This CMG method has been well-validated against electronic pill cap monitoring, serum/urine drug levels, physiological drug effects, change in clinical control (such as blood pressure measures), co-morbidity, and cost [16, 18,19,20]. In addition to health systems like the Veterans Administration (VA) and Kaiser Permanente, CMG has also demonstrated acceptable inclusiveness and validity in diverse, low-income safety net populations [21]. This analysis is limited to respondents who had a calculable adherence measure.

CMG adherence for three medication groups (typically treating the three chronic conditions of patients in the study) was examined: (1) oral hypoglycemics, (2) anti-hypertensives, and (3) lipid-lowering drugs. Adherence was defined as average adherence for each prescribed medication, weighted by the number of days within each observation window for each medication (i.e., time between first and last fill). The four adherence measures are: “overall adherence” (adherence rate across all medications) + an adherence measure rate for medication(s) specific to the aforementioned groups “DM-med adherence” (oral hypoglycemics), (2) “HTN-med adherence” (anti-hypertensives), and “HL-med adherence” (lipid-lowering drugs). In addition, dichotomous versions of all CMG adherence measures were created to describe patients as “highly adherent” or “poorly adherent.” Patients whose reverse-coded CMGs were less than 80% (i.e., gap in therapy>20%) were classified as “poorly adherent,” whereas those greater than 80% were classified as “highly adherent,” a determination that is based on prior studies [16,17,18,19,20].

Predictor Variables

Patients were provided with a brief description of the Internet refill system and then asked, “Have you used the health systems Internet site to refill any medications in the last 12 months?” Respondents who answered “do not know” (n = 14 for Internet) were categorized as “no.” LEP status was captured by electronic medical record data—patient self-identification of a primary language other than English, plus a claims record of use of an interpreter.

Other Measures

Other survey measures were age, gender, marital status, race/ethnicity, language/dialect, education, household income, years in the USA for foreign-born patients, Medicare/Medicaid insurance status, health status, chronic condition, number of prescribed medications, and social support (eight questions gauging resources; final score of 0 “no support” to 40 “always supported”). Missing responses were less than 1% for all the variables except the income question, where 4.9% refused to answer and 8.6% answered “do not know.”

Analysis

All statistical analyses were conducted using STATA version 14.2 (College Station, TX) software, and a p value of < 0.05 was used to determine statistical significance. We calculated inverse probability weights (IPWs) to account for differential survey response from study participants by age group and language. The IPWs were applied in the analysis to mitigate bias incurred by survey non-response. The probability modeling using IPWs for this cross-sectional survey has been described previously [6].

Univariate summary statistics, bivariate chi-square analysis on categorical variables, and t tests and one-way analysis of variance on continuous variables were done for cases with calculable CMGs. We compared these across LEP and EP cases. We used the literature on adherence and limited English proficiency and significant associations in bivariate analyses to determine the final set of covariates, given the sample size. We used multivariate linear regression models to examine the adjusted effects of use of the Internet refill system on overall medication adherence and compared this association between LEP and EP patients using stratified models. Final models were adjusted for age, gender, education, insurance, chronic conditions, and number of medications.

Results

A total of 509 patients completed the survey. The overall response rate for the survey was 35.5% (n = 509; 34.5% LEP, n = 328; and 37.4% EP, n = 181). This analysis is limited to the 384/509 (75%) who had a calculable CMG adherence measure—134/181 (74%) EP and 250/328 (76%) LEP. Respondents who had calculable CMGs were older (66 vs. 63 years, p = 0.001), on more prescribed medications (6 vs. 3, p < 0.001), and were more likely to be on Medicare/Medicaid insurance (31 vs. 18%, p < 0.001) versus those who did not have a CMG adherence measure (n = 125) (data not shown). There were no other significant differences across all characteristics, even within the two language proficiency groups.

The weighted distributions and means for patient demographics and health characteristics for the 384 participants with a CMG adherence measure, by LEP status, are shown in Table 1. Compared with EP respondents, LEP individuals were younger, more likely to be married, less college/university educated, and had lower self-rated health status. The LEP group had a higher percentage of diabetes but no difference in hypertension or hyperlipidemia diagnoses. LEP patients also had a lower number of prescribed medications and a lower social support score versus the EP patients. Notably, LEP patients had significantly lower use of Internet remote refill systems compared to the EP patients (22 vs. 56%, p < 0.001).

Table 1 Weighted percentages and means for patient characteristics by English proficiency status among respondents with a calculable CMG (n = 384)
Full size table

Among all patients (Table 2), overall medication adherence was high, and LEP patients had a lower overall adherence rate versus the EP group (0.901 vs. 0.938, p < 0.001) and a lower percentage of being “highly adherent,” defined as adherence > 0.80 (87 vs. 96%, p < 0.001). Furthermore, LEP respondents had lower medication adherence and lower percentages of highly adherent for all three condition-specific medication categories.

Table 2 Adherence outcomes across English proficiency status
Full size table

The multivariate linear regression analysis for overall adherence among all patients and in stratified models (LEP alone, EP alone) is shown in Table 3. LEP status (β = − 0.022, p = .047) and a diagnosis of hyperlipidemia (β = − 0.025, p = .008) were negatively associated with overall adherence. Use of the Internet refill system (β = 0.030, p = .002), Medicare or Medicaid insurance (β = 0.030, p = .007), and a college/university education (or higher level) (β = 0.0029, p = .029) were positively associated with adherence. The interaction term (LEP*Internet refill use) was not significant (p = 0.89). In stratified models, use of remote Internet refills was positively associated with adherence, even when examining LEP (β = 0.029, p = .003) and EP patients (β = 0.027, p = .049) separately. In the EP model of overall adherence, Medicare/Medicaid insurance status was positively associated (β = 0.040, p = .01), while among LEP patients, hyperlipidemia was negatively associated with adherence (β = − 0.035, p = .004).

Table 3 Multivariate regression analysis of overall adherence and Internet medication refills among the n = 384 patients with calculable CMGs, and in stratified models (EP vs. LEP)
Full size table

Although the sensitivity tests are not shown, we also compared LEP Internet refill users and LEP non-users across all variables used in this analysis and across some survey questions related to English language understanding. There were no significant differences for these, within the LEP group, across Internet refill user status. For the same comparison of study variables among the EP patients alone, we observed that Internet refill users were significantly younger and had higher percentages of patients with household income > 75 K, and also a university/college level education.

Discussion

We found that the use of remote medication refills through an Internet portal was independently and significantly associated with higher medication adherence in both EP and LEP patients. Compared to EP patients, LEP status was associated with lower use of Internet portal refills and decreased medication adherence even after adjusting for covariates. We add to the literature by demonstrating that linguistically diverse patients are under-utilizing a portal tool that is positively associated with medication adherence. Our results extend previous studies that focused on race/ethnicity (but not language) and the relationship between Internet portal medication refills and medication adherence [2, 5, 7, 15].

Some strengths of this study include the linguistically diverse cohort and the use of the well-validated CMG to measure adherence. CMG does not measure actual medication taken—it is based on time only, which is a known limit of this measure. And while only 75% of respondents had a calculable CMG, this level is equivalent and/or higher than the CMG calculability rates achieved in other comparable populations [21]. Another related limitation is that CMG was calculated for three primary groups of ongoing medication prescriptions and did not include medications for other common chronic conditions (such as COPD or asthma) or adherence to new prescriptions. However, the study intentionally focused on patients with one of three chronic medical conditions (DM, HTN, HL) that would probably be treated with the medication classes included. There were adequate rates of calculable CMGs across patients with each of the three ICD-9 diagnoses, confirming that no group of patients were “left out”: 82% of patients with a DM diagnosis, 72% of patients with HL patients, and 85% with HTN patients had a calculable CMG for at least one medication. It is also possible that our adherence measure did not capture cases of patients taking the aforementioned medication groups through another health system or in another pharmacy by paying “cash only.” This specific limitation may be better addressed in future studies by also asking patients if they get any medications outside of their primary health system.

The Group Health Internet portal was offered in English during the time of the study, and so, even LEP patients utilized the refill system in English. This poses the question as to whether LEP patients using the Internet refill system had better English proficiency versus non-users or were different in any other significant ways. The telephone survey administered to all patients included questions from the Consumer Assessment of Healthcare Providers and Systems Survey (CAHPS), some of which are related to English understanding/communication. As a sensitivity test, we queried whether English language variability as assessed by these questions, accounted for differences in portal use among the LEP patients. We found no such differences between LEP Internet refill users and LEP non-users. Among the EP Internet refill users, sociodemographic variables such as younger age, higher education, and income level were more predictive of Internet refill use (as expected). Interestingly enough, this was not the case among LEP patients. These findings highlight that there may be very different predictive factors for uptake and use of these Internet portals for LEP patients and should thus be investigated more thoroughly. Future work must examine the related factors (other than English proficiency), such as cultural differences, which account for the differences in portal uptake between English-speaking and non-English-speaking patients. Additional studies among diverse populations should also be conducted with multilingual portal options, to better gage the effect of these tools in the patient’s preferred real-world setting.

The cross-sectional design does not allow for inference of casual relationships. Given that 509 of 1490 invited patients completed the survey, this does limit the representativeness of the final analytic sample. In addition, we used self-reports for certain measures, which are subject to recall bias and socially desirable answers. Of note, our main predictor (Internet portal use for refills) was self-report and so does not account for actual use. Furthermore, we did not distinguish between frequent or occasional use of the Internet patient portal and its impact on medication adherence. Prior studies have shown that there may be a dose response effect between Internet portal medication refills and medication adherence [2]. Our results may not generalize to all patients with LEP, those that speak other languages, or other healthcare systems.

Finally, adherence rates for ongoing prescriptions were good across all patients in the analysis—and these levels are similar to adherence rates observed for patients in comparable settings like Kaiser Permanente [2]. The important finding, nonetheless, is that a medication adherence rate gap exists between LEP and EP patients, even in a population that has access to high quality of health care. It is likely that linguistic disparities in use of Internet refill systems and adherence are more pronounced in underserved settings (outside of systems like GHC and KP). The study also took place before the full expansion of the Affordable Care Act (ACA), as large groups of vulnerable populations came under commercial insurance plans like GHC and KP. Therefore, examining the medication adherence gaps in LEP populations today in these commercial systems, and the effect of Internet patient portal use, is necessary, as these populations are insured.

These findings suggest that LEP patients with chronic conditions are at an increased risk of sub-optimal adherence and are under-utilizing an Internet portal tool that is significantly associated with better medication adherence. The adoption of digital technology is accelerating rapidly, and the Internet has become a now ubiquitous interface for healthcare-related transactions and communication. Should our healthcare system fail to provide LEP patients with the structural access, user-friendly and culturally appropriate digital interfaces, knowledge, and the confidence to digitally access services through an Internet portal, we face the risk of widening existing healthcare gaps among these already-vulnerable populations.