Design, Setting and Participants
The study was a cross-sectional analysis of baseline survey responses of participants from two sites in a study of language barriers and communication. We recruited patients from the Alameda County Medical Center cardiology clinic and inpatient general medical-surgical ward with the goal of obtaining a Spanish-speaking Latino sample and an English-speaking comparison group of any ethnicity. Recruitment occurred between March 2004 and August 2005. Bilingual research assistants identified Spanish-speaking patients by surname and primary language as it was listed on the clinic schedule or on the inpatient board and registration card, and by consulting nursing staff at both locations. Research assistants then confirmed the patient’s primary language at the time of enrollment. On the days that interviewers were on site, efforts were made to recruit all patients who appeared to be primarily Spanish speakers; the days of the week on which recruitment occurred were varied to include a representative sample of LEP patients. For purposes of comparison, English speakers were recruited at both sites over the same time period as the Spanish speakers.
Informed consent was obtained from patients prior to their baseline interview. Individuals were excluded if they were younger than 18 years old, spoke a language other than English or Spanish, or failed a cognitive screen at the beginning of the interview (Mini-Cog).23–26 Research assistants conducted a face-to-face interview with the participants in their language of choice. The institutional review boards of the University of California San Francisco and Alameda County Medical Center approved the protocol.
Outcomes and Follow-up
Demographic variables included age, gender, and self-reported ethnicity. Three language-related variables were used in this analysis: English proficiency, language preference for medical care, and potential benefit derived from language assistance services. First, to assess English proficiency, participants were asked the Census-LEP item: “How well do you speak English?” Response options were “not at all, not well, well, or very well”. Second, to assess language preference, participants were asked, “In general, in what language do you prefer to receive your medical care?” Response options were “English, Spanish, or both equally”. The language preference item was dichotomized into English and English and Spanish equally, versus Spanish. Finally, we created the composite variable “benefit from language assistance” by combining the Census-LEP with the language preference for medical care question as described in the Figure. This new composite variable “benefit from language assistance” was dichotomous, with individuals categorized as likely to benefit (benefit group) versus unlikely to benefit (no benefit group) from language assistance. Persons answering “not at all” or “not well” to the Census-LEP were included in the “benefit” group; in addition, those answering “well” but indicating their language preference as Spanish, were also categorized in the “benefit” group. Those patients either answering “very well”, or answering “well” to the Census-LEP but reporting their language preference as English or both English and Spanish equally, were categorized as “no benefit.”
We assessed participants’ self-reported ability to communicate effectively in English with their physicians (outcome variables) using the questions: “How well can you discuss your symptoms with your doctors in English?” and, “How well can you understand your doctors’ recommendations in English?” These questions were asked immediately preceding the visit for the cardiology clinic patients, and during hospitalization for the inpatients. Both communication questions had the same response options as the Census-LEP question. As used in prior research on physician communication,27 each of these questions was dichotomized, such that only those answering “well” or “very well” were considered able to communicate effectively in English with their physicians.
Statistical analyses were performed using STATA 9.28 We used receiver operating curves (ROC)29–31 to assess the value of two different response thresholds of Census-LEP—a high threshold of less than “very well” (LEP) vs. “very well” (English-proficient), and a mid-threshold of less than “well” (LEP) vs. greater than or equal to “well” (English-proficient)—in predicting effective communication with physicians. Thus, the high threshold definition of LEP included those who responded “well” to the U.S. Census question, and the mid-threshold did not. Using the two outcomes of patient-reported ability to discuss symptoms and to understand physician recommendations in English as the gold standard, we compared the sensitivity and specificity of the following: 1) each of these two response thresholds of the Census-LEP question alone; 2) the preferred language for medical care question alone; and 3) the new composite variable “benefit from language assistance”.