Our research used two data sources: the Medical Expenditure Panel Survey (MEPS) [22] and the National Health Interview Survey (NHIS) [23]. The sampling frame of MEPS is drawn from the participants in NHIS. Immigration information is not contained in MEPS, so we used the NHIS data to identify immigrant workers in MEPS. We pooled and linked data from the 2003–2007 NHIS and the 2004–2009 MEPS to construct a database to investigate medical expenditures for nonfatal occupational injuries in the U.S.
The MEPS is conducted annually and cosponsored by the Agency for Healthcare Research and Quality (AHRQ) and the National Center for Health Statistics (NCHS). It provides nationally representative estimates of health care use, insurance coverage, medical expenditures and sources of payment for the civilian non-institutionalized population. The MEPS has two major components: the household component (MEPS-HC) and the insurance component (MEPS IC). MEPS-HC obtains data from a nationally representative sample of households through an overlapping panel design in which new respondents are sampled and recruited from NHIS respondents each year and are interviewed 5 times over a 2.5-year period. Respondents are questioned about medical expenditures incurred in a 2 year reference period. This provides continuous and current estimates of health care expenditures at both the person and household level for each calendar year. An additional component of MEPS, the medical provider component (MPC) supplements and corroborates information received from the MEPS-HC component; the information from the MPC is incorporated into the MEPS-HC data [22]. MEPS-IC is an annual survey of employers that collects information on the employer's health insurance offerings.
In the MEPS, total expenditures were defined as the sum of payments paid for medical care services, including out-of-pocket payments, payments from private insurance, payments made by Medicare and Medicaid, or payment by workers' compensation, or other sources. Payments for over-the-counter medications and for alternative medicine (e.g., acupuncture or chiropractic care) are not included. The AHRQ applies imputation methods using available charge and payment data in either the MEPS Household Component or the MEPS Medical Provider Component to replace missing expenditure data [24].
Human participant protection
The data were collected with the informed consent of the respondents of the NHIS, following procedures approved by the Institutional Review Board of the National Center for Health Statistics. The institutional review board of the Research Institute at Nationwide Children's Hospital approved secondary analysis of the data for our study.
Terms and definitions
Immigration status and workers
To determine the immigration status of respondents, responses to the question "Where were you born?" in the NHIS were used. A respondent was categorized as being immigrant if the birthplace was outside the U.S. The NHIS categorized respondents born in U.S. territories as foreign-born because they may have a culture different from mainstream U.S. culture and because respondents in U.S. territories come from more than one "culture." Therefore, in our analysis U.S.-born workers included only those individuals born in one of the 50 states or the District of Columbia. Less than 1% of respondents of MEPS did not report their birthplace. Our analysis indicated that there was no significant difference in sociodemographic characteristics (gender, age, education level, family poverty level, and having no medical insurance) between respondents with birthplace and those without birthplace information.
Workers were defined as those who self-reported employment in any round of the 5 MEPS interviews. Some workers did not finish all five rounds so we calculated the number of follow-up days that these workers participated in the MEPS survey (Figure 1).
Occupational injuries
As described above, MEPS uses five rounds of interviews to collect detailed data on medical conditions, health care use, medical expenditures, sources of payment, and health insurance coverage for a 2 year reference period. When a medical condition is first reported in the MEPS, a portion of the MEPS Household Component questionnaire asks the respondent to specifically report whether this medical condition is an injury or not (yes or no). When an individual 16 years of age or older reports an injury in the MEPS, a number of questions are asked, including "Whether the injury occurred at work." In our study, injuries were defined as occupational injuries if the respondent said that the injury occurred at work. A respondent could have reported multiple injury events but each injury event has its own date of injury and event identification number.
In the 2004–2009 MEPS datasets, we found that some injuries occurred many years prior to the MEPS interview, but still had associated medical expenditures. Because our study aimed to investigate the medical expenditures of acute injuries that occurred during the MEPS reference period, we excluded a total of 140 injuries that occurred prior to the first MEPS interview. In this study, if no medical services were reported, we assumed medical care was not sought for that particular injury.
Health insurance coverage
The MEPS-IC collects data on health insurance plans obtained through private and public-sector employers. Data are obtained from employers through a prescreening telephone interview, a mailed questionnaire, and a telephone follow-up of non-respondents. In this article, we used the summary health insurance coverage variable in the MEPS for each respondent to categorize respondents as having private insurance, public insurance, or no insurance. The private insurance category includes respondents who, at any time in the survey year, had individual or group plan coverage for medical or related expenses or who were covered by TRICARE, the Department of Defense heath care program. Public insurance includes respondents who were never covered by private insurance or TRICARE during the year but who were covered at any time by Medicare, Medicaid, SCHIP (State Children's Health Insurance Program), or other state and local medical assistance programs. Those in the no insurance coverage category include respondents who did not have private or public insurance coverage at any time in the calendar year. This health insurance classification was used by the AHRQ to describe health insurance status of full-time U.S. workers in a recent report [25].
Statistical analyses
Data analyses were conducted using SAS [26]. All medical expenditures were adjusted to 2009 equivalent dollars using the Consumer Price Index for Medical Services published by the U.S. Bureau of Labor Statistics [27]. Due to the fact that multiple years of MEPS data were pooled together, we adjusted the final weight variable before our statistical analyses. AHRQ has recommended adjusting the analytic weight variable by dividing it by the number of years being pooled. This adjustment would have no effect on estimated means, proportions, or regression coefficients because the weight variable was being divided by a constant (i.e., number of years).
We first identified the total number of workers who reported occupational injuries, the total number of occupational injuries, and the total number of occupational injuries for which medical services were sought (those injuries with non-zero medical expenditures) in the MEPS sample. Using the survey design variables and the adjusted final weighting variable, we provide national estimates of the total number of injured workers, the total number of occupational injuries, and the total number of occupational injuries for which medical services were sought during the 2 year reference period. Using the follow-up days that each worker participated in the MEPS interviews and the total number of occupational injuries, we calculated an annual incidence rate and 95% confidence intervals (CI) of occupational injuries per 100 workers. We also calculated the proportion (%) and 95% confidence intervals (CI) of occupational injuries for which medical services were sought. We used the bootstrap methods with 1000 repeating bootstrap samples to calculate the 95% CIs [28].
Second, in order to test the hypothesis that immigrant workers were less likely than U.S.-born workers to seek medical services after occupational injuries, we used a hierarchical logistic regression modeling approach in which immigrant status and sociodemographic variables were treated as independent variables while seeking medical treatment was treated as the dependent variable. We included sociodemographic variables gender, age, race/ethnicity, marital status, education level, family poverty status based on federal poverty levels (FPL), and health insurance coverage of persons who reported occupational injuries. These variables are often adjusted as important factors in medical expenditures/costs studies [19, 20, 29].
Third, we provided national estimates of the average expenditures of occupational injuries according to the type of medical services and sources of payments and calculated the proportion (%) and 95% CIs of different types and sources of payment in the total expenditures during the 2-year MEPS reference period. Bootstrap methods were again used to calculate the 95% CIs. Results regarding the sources of payment allowed us to test the study hypothesis that the proportion of medical expenditures paid by workers’ compensation for occupational injuries is smaller for immigrant workers than for U.S.-born workers.
Finally, to estimate the average medical expenditures for occupational injuries per injured worker during the MEPS reference period and to assess the impact of above-mentioned sociodemographic variables on medical expenditures, we used multivariable linear regression models that have been used by others to analyze medical cost data [19, 20, 30]. Since the expenditure data were right skewed, a natural logarithm transformation was used to transform the expenditure data. Then we used linear regression to examine the association of the log-transformed expenditures with immigrant status, adjusting for sociodemographic variables. A smearing factor was applied to generate the final mean expenditures using Duan's approach to adjust for the impact of the log-transformation [31].