To create cohorts of patient-provider language concordance, we combined responses to the questions “What language is spoken in your home most of the time?” and “Does someone at your provider’s speak the language you prefer or provide translator services?” If English was spoken at home, the subject was categorized as English-Concordant. If English was not spoken at home and someone at the provider’s spoke the respondent’s preferred language or offered translation services, the subject was categorized as Other Language-Concordant. Subjects who reported not speaking English at home and denied that someone at their provider’s spoke their preferred language or offered translation services formed the third cohort, Other Language-Discordant (Fig. ).
We assessed CRC screening using self-reported rates of FOBT and endoscopy. Given that patients may not have FOBT exactly within 12 months, we considered tests performed within 2 years prior to the date of MEPS survey completion to be current with recommendations. In MEPS data, responses for sigmoidoscopy and colonoscopy were combined into a single variable with time frame choices either within or greater than 5 years of the survey date. Therefore, if an individual had FOBT within 2 years or endoscopy within 5 years, they were classified as being current with CRC screening.
Covariates included: self-reported race/ethnicity, age, education, marital status, family income, employment status, time since last check-up, and health insurance status. Race and ethnicity were combined into a variable with five categories: white non-Hispanic, black, Hispanic, Asian, and other. Age had three categories: age 50–<65 years, 65–<75 years, and age 75–85 years (this category may include subjects >85 years as MEPS top coded ages at 85). Education had four categories: no degree, high school or equivalent, some college or greater, or other. Marital status had two categories: married or not married. Total family income had four categories defined by the federal poverty line (FPL): poor/near-poor (<125% FPL), low income (125–<200% FPL), middle income (200–<400% FPL), and high income (≥400% FPL). Employment status had two categories: employed or unemployed. Time since the last checkup had three categories: within 2 years, greater than 2 years, and never. Health insurance status had three categories: any private insurance, public insurance only, and no insurance.
Co-morbidities are often considered when recommending CRC screening; however, these indices are not included in MEPS. As a proxy, the Physical Component Score (PCS) and Mental Component Score (MCS) of the 12 Item Short Form Health Survey (SF-12) were used. These scores have been well validated and are standardized with a mean of 50 for the general population.24
Scores were converted into two categories: scores less than the sample median (“low” PCS/MCS) and those greater than or equal to the median (“high” PCS/MCS). Sample PCS and MCS medians were 47 and 54, respectively. Of note, Medicare coverage of endoscopy for average-risk adults began in 2001, so year of survey was also included. We included US region to account for regional variations. The following provider-level variables were also included: provider race, ethnicity, and sex; and provider type and specialty.
To account for the complex sample design, survey statistical procedures were used. Weighted prevalence and standard error (SE) estimates were calculated for independent variables using MEPS survey weights, and χ2
tests assessed for differences between cohorts. Variables with proportion of missing responses greater than 65% (provider characteristics) were eliminated. For the remaining variables, individuals with complete data were compared to those with missing data to assess generalizability. All subsequent analyses were done on samples with complete data for all variables retained (n
Bivariate odds ratios (ORs) and 95% confidence intervals (CIs) were used to evaluate the associations between each independent variable and receipt of CRC screening. Variables were independently assessed for confounding or effect modification of language concordance. Those yielding a ≥10% change in the magnitude of effect were considered as potential confounders.
We used multivariate logistic regression to determine the association of language concordance with CRC screening. We included in the model those variables determined a priori to be potentially associated with screening (sex, age, time since last checkup, marital status, employment status, year, region), as well as variables found to be confounders (race/ethnicity, education, family income, and health insurance status) or effect modifiers (PCS).
We evaluated our definitions of both the primary explanatory variable and outcome variables. Patient language was re-defined by comfort level speaking English using the question “Are you comfortable conversing in English?” Individuals with LEP (those who responded “No”) were then grouped by whether someone at their provider’s spoke their preferred language or offered translation services. In this way, three cohorts based on English proficiency were created: English-proficient, LEP-Concordant and LEP-Discordant. A simple logistic regression using these re-defined cohorts was compared to that using the original cohorts based on language spoken at home. In addition, receipt of CRC screening was re-defined in several ways. We assessed receipt of FOBT alone, endoscopy alone, endoscopy ever, and any screen ever.
All prevalence, odds ratio, and variance estimates are presented from weighted analyses unless otherwise specified. Statistical significance was set at α
0.05. All analyses were conducted with SAS (version 9.2, SAS Institute Inc., Cary, NC). This study was granted exempt status by the Boston University Institutional Review Board.