The study recruited patients between July and December 2025 at Boston Medical Center, a tertiary care academic medical center that serves as New England’s largest safety-net hospital. Half of its patients identify as Black and 25 percent as Latino, and more than one-third prefer to receive medical care in a language other than English, with Spanish accounting for 54 percent and Haitian Creole for 18 percent of those discharged from medical hospitalizations. Researchers approached 240 eligible patients, of whom 115 expressed interest; after exclusions during teach-back assessment, consent, and survey completion, the final analytic sample included 89 participants. The mean age was 53.7 years, with a slight female predominance at 52.8 percent. Nearly all participants, 95.5 percent, reported speaking English less than “very well,” meeting the broad definition of limited English proficiency, and fewer than half, 43.8 percent, reported reading “very well” in their preferred language.

The researchers chose in-person recruitment of hospitalized patients rather than the online panels often used in discrete choice experiments for two reasons: individuals with non-English language preference often face technology and language barriers that reduce participation in online panels and may bias samples, and recruiting during hospitalization helps participants engage with discharge-related trade-offs. Language-concordant research assistants approached eligible patients, excluding those on infectious or suicide precautions or lacking consent capacity as reported by nurses. The consent process included teach-back to ensure understanding, and during the experiment itself participants could either hold the tablet themselves or have the research assistant read the choices aloud while displaying accompanying images, a flexible approach designed to accommodate varying levels of technological familiarity and reading comfort.

Discrete choice experiments work by presenting patients with pairs of hypothetical discharge scenarios that vary across attributes and asking them to choose between options. Each participant completed 20 unlabeled choice sets, yielding 40 observations per participant. Because a full factorial design crossing all attributes and levels would have produced 108 possible combinations, the team used a fractional factorial design generated in Sawtooth Software, which preserves orthogonality and balance while minimizing respondent burden. From the observed choices, researchers estimated part-worth utilities, numerical values representing the relative desirability of each level within an attribute, using a mixed logit model that accounts for the panel structure of the data and adjusts for heterogeneity across the sample. Relative attribute importance scores were calculated as the range of utilities within each attribute divided by the sum of ranges across all attributes, expressed as percentages, with 95 percent confidence intervals produced by parametric bootstrapping.

The results show that record of discharge instructions carried a relative importance of 39.9 percent (95 percent CI: 34.0 to 44.5 percent) for the combined sample, followed closely by verbal communication at 34.7 percent (95 percent CI: 29.0 to 39.1 percent). Family involvement contributed roughly 21 percent, while post-discharge phone contact was the least important attribute by far, at just 4.8 percent (95 percent CI: 1.4 to 10.5 percent). Preferences, however, differed by language group. Spanish speakers placed the highest value on the form of discharge instructions at 45.1 percent (95 percent CI: 37.7 to 51.6 percent), while verbal communication was most important to Haitian Creole speakers at 48.9 percent (95 percent CI: 37.0 to 58.3 percent).

Within the discharge instructions attribute, the strongest preference across both language groups was for instructions written in the patient’s preferred language, with a part-worth utility of β = 1.26 (p < 0.001) compared with instructions written in English, the reference condition. Audio-recorded instructions in the preferred language were also strongly preferred over English-written instructions (β = 0.72, p < 0.001), though less so than written instructions. For verbal communication, both groups showed a marked preference for language-concordant providers over phone interpretation (Spanish: β = 0.85; Haitian Creole: β = 1.13; both p < 0.001). Spanish speakers preferred in-person interpreters over phone interpreters, but this difference was not statistically significant (β = 0.27, p = 0.062). Haitian Creole speakers showed no significant difference between in-person and phone interpretation (β = −0.09, p = 0.667), but viewed video interpretation negatively compared with phone interpretation (β = −0.40, p = 0.046).

Family involvement in discharge teaching also shaped preferences. Participants overall favored having family present in person for discharge teaching (β = 0.65, p < 0.001). Haitian Creole speakers also expressed a significant preference for family involvement by phone compared with no family involvement (β = 0.66, p < 0.001), but Spanish speakers did not (β = −0.08, p = 0.497). For post-discharge phone contact, there were no significant differences between receiving a call from a hospital nurse (β = 0.15, p = 0.120) or a primary care nurse (β = 0.14, p = 0.148) compared with having a number to call if questions arose, a pattern consistent across both language groups.

Exploratory subgroup analyses revealed meaningful heterogeneity within patient populations. Participants with low English fluency, defined as speaking English “not at all” or “not well,” valued recorded discharge instructions more (41.0 percent, 95 percent CI: 35.0 to 46.2 percent) than verbal communication (35.5 percent, 95 percent CI: 28.9 to 40.8 percent). In contrast, those with low reading ability prioritized verbal communication (42.7 percent, 95 percent CI: 34.8 to 47.8 percent) over written discharge instructions (29.3 percent, 95 percent CI: 21.8 to 34.4 percent). Language-concordant providers were strongly preferred, especially by participants with low reading ability (β = 1.29, p < 0.001). Written instructions in the preferred language showed similarly strong preference among those with five or fewer years in the United States (β = 1.58, p < 0.001) and those with low reading ability (β = 1.32, p < 0.001), exceeding audio-recorded instructions in both groups. Family visiting in person was favored by participants with more than five years in the United States (β = 0.75, p < 0.001) and those with low reading ability (β = 1.01, p < 0.001), whereas among those with five or fewer years in the country, family visiting in person was significantly less preferred than no family involvement (part-worth utility: −0.59, 95 percent CI: −1.05 to −0.12).

Choice simulations estimated how likely patients would be to select specific discharge profiles compared with reference conditions. A profile with family present in person was chosen 60.2 percent of the time; one with a language-concordant provider, 69.5 percent; audio-recorded discharge instructions in the preferred language, 66.7 percent; and written instructions in the preferred language, 76.6 percent. When all four preferred features were combined into a single profile, the selection probability rose to 87.0 percent. These simulations translate the statistical estimates into concrete predictions about how patients would respond to different discharge designs, offering health systems a way to anticipate the uptake of specific intervention components.

The authors note that the findings address a persistent evidence gap. Language barriers exacerbate miscommunications during hospital discharge, contributing to preventable readmissions and adverse outcomes, yet patients with non-English language preference remain systematically underrepresented in discharge intervention research, with most trials excluding them. Prior work has documented the value of language-concordant providers and professional interpreters, but little research has quantified how patients weigh competing discharge features against one another. The study advances prior research by demonstrating that the record of discharge communication, whether written or audio-recorded in the preferred language, is the primary driver of preferences, whereas earlier studies had emphasized verbal communication alone. The preference for instructions patients can review after leaving the hospital aligns with evidence that written materials improve comprehension and retention better than verbal explanations, and the authors suggest this may reflect the cognitive demands of hospitalization itself, when illness severity can limit verbal processing.

The implications for health systems are substantial, though implementing changes will require operational investment. Few health systems systematically incorporate language preference into discharge processes, and the authors suggest the greatest gains will come from translating written instructions and providing verbal education through language-concordant staff. Challenges remain, including the limited availability of bilingual providers and the cost and logistics of real-time professional translation in busy inpatient settings. Emerging artificial intelligence translation tools offer promise but still require human verification under current regulations, limiting immediate use. Although audio-recorded instructions were less preferred than written ones, they may offer a practical complement after discharge, and telehealth models may enable flexible deployment of multilingual clinicians for verbal review.

The study has limitations that constrain generalizability. The single-site design at one academic safety-net hospital limits the extent to which findings apply to other settings, and the modest sample size of 89 participants constrains statistical power, particularly for the exploratory subgroup analyses, which lacked adequate sample size for formal statistical testing. Reading ability and English fluency were self-reported rather than formally assessed, and individuals with immigration-related concerns may have been less willing to participate, potentially introducing selection bias. The researchers also did not systematically document whether participants read and selected choice sets independently or had them read aloud by a research assistant, which may have influenced selections. Notably, the observed preference differences between and within Spanish and Haitian Creole speakers suggest that language alone inadequately predicts patient priorities, and the authors argue that future work should develop scalable, adaptive methods to personalize discharge interventions to individual linguistic, cultural, and social preferences rather than relying on one-size-fits-all approaches, and should examine whether implementing these preferred aspects of discharge care results in better patient understanding, reduced adverse events, and more equitable post-discharge outcomes.

Subject of Research: Patient preferences for hospital discharge communication among patients with non-English language preference

Article Title: A Discrete Choice Experiment to Understand Hospital Discharge Preferences of Populations with Non-English Language Preference

Article References: Jain, A., Lugo, N., Bautista-Hurtado, D., Zuniga, S., Penugonda, A., Philippe, N. J., Labossiere, S., Pluta, K., Cabral, H., Mitchell, S., Jack, B. W., Fernandez, A., Drainoni, M.-L., Salloum, R. G., & Austad, K. E. (2026). A Discrete Choice Experiment to Understand Hospital Discharge Preferences of Populations with Non-English Language Preference. Journal of General Internal Medicine. https://doi.org/10.1007/s11606-026-10921-6

Image Credits: AI Generated

DOI: 10.1007/s11606-026-10921-6

Keywords: discrete choice experiment, limited English proficiency, hospital discharge, language barriers, patient preferences, health equity, interpreter services, readmissions, safety-net hospital, transitional care, Spanish speakers, Haitian Creole speakers