Subjects and Setting This is a retrospective cohort study using administrative data of all non-disabled Medicaid beneficiaries age 35 years and older discharged from all acute care California hospitals with a diagnosis of AMI from 1994 to 1998. These older data were chosen to allow for comparison with findings in the aforementioned Canadian study. Data from individuals under 35 years of age were excluded from the analysis in an effort to isolate AMI secondary to coronary artery disease rather than other causes such as congenital disorders.
The data for this study come from two main sources: (1) the annual California hospital discharge data available from the California Office of Statewide Health Planning and Development (OSHPD), and (2) the California Department of Health Services (DHS) Medicaid monthly eligibility file. The California hospital discharge record includes, among other things, information on admission date, discharge date, patient demographics, and diagnosis codes. OSHPD applies several hundred audit rules to ensure the validity of their data before making them available. Data elements are not released if they exceed an error tolerance level of 0.1%.17
OSHPD does not, however, require reporting of individual language preference. Therefore, we used a special research file that linked hospital discharge data from OSHPD with the DHS Medicaid eligibility file for the period 1994–1998; language preference is routinely collected by Medicaid.
Primary Predictor Language preference was entered into the DHS Medicaid eligibility file at the time of enrollment. Individuals with English entered in the language field were considered to have an English preference (EP). All other languages were considered to be a non-English preference (NEP).
Outcome Variables We examined 2 outcomes in our analyses: length of stay (LOS) and in-hospital mortality. LOS was logarithmically transformed to normalize the distribution. Data from hospital LOS of 0 were excluded from analysis as these records were thought to represent events of early death or transfer to another facility, rather than effects of language preference. Among these excluded records, there was no difference in hospital mortality by language preference.
We adapted demographics and health status covariates from a previously validated prediction model of 30-day mortality after hospitalization for AMI.18
Specifically, our risk adjustment model included 24 covariates: race/ethnicity (white, black, Latino, and Asian/other), age (by quartile: 35–49, 50–64, 65–79, and 80 and older), gender, year of admission, acute renal disease, catastrophic sequelae of AMI, anterior wall infarction, congestive heart failure, high-risk malignant neoplasm, hypertension, inferior wall infarction, pulmonary edema, sepsis, shock, end stage renal disease, CNS disease, complete atrioventricular block, complicated diabetes, other cerebrovascular disease, paroxysmal ventricular tachycardia, history of prior coronary bypass surgery, prior angioplasty procedure, chronic obstructive pulmonary disease, and acute anemia. The validated prediction model also included insurance type, number of prior admissions, and resuscitation status. We did not adjust for insurance status as all individuals included in the study had 1 type of insurance (Medicaid). We did not adjust for number of prior admissions because the dataset did not include reliable unique identifiers to determine if records represented repeat admissions for a given individual. We did not adjust for resuscitation status as OSHPD did not record this variable until 1999.Procedure codes from hospital discharge data were used to create a variable for cardiac procedure. Records were categorized as either having no cardiac procedure or surgery; a cardiac procedure (catheterization, coronary artery stent placement, or angioplasty); or cardiac surgery (coronary bypass grafting).
Analysis We used multivariate regression step-wise modeling to explore whether observed differences between the LOS and in-hospital mortality between non-English preference (NEP) and English preference (EP) could be explained by patient demographics/health status, having a cardiac procedure/surgery, or hospital of care. We began our linear regression modeling with an unadjusted analysis of language preference as a predictor of logarithmically transformed LOS. Second, to control for the effect of patient level characteristics, we added the variables from OSHPD’s validated risk adjustment model. Third, to control for the effect of treatment differences, we added variables for receipt of cardiac procedure or surgery. Finally, to control for effects specific to individual hospitals, we added an indicator variable for each hospital to the model. We calculated the regression coefficient for percentage increase in length of stay for NEP as compared to EP individuals.We repeated this step-wise modeling for in-hospital mortality using multivariate logistic regression. We calculated odds ratios of NEP in-hospital mortality as compared to EP in-hospital mortality.