Study Design
This was a retrospective cross-sectional study using the Medical Expenditure Panel Survey (MEPS) data of 2012, 2014, and 2016. MEPS is a nationally representative survey conducted by the Agency for Healthcare Research and Quality (AHRQ) of the US noninstitutionalized civilian population (10). MEPS has information on demographics, socioeconomic characteristics, medical conditions, health status, and other health-related data. The International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes was used to report the medical conditions. Beginning in 2016, the International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) codes used to report medical conditions.
Study Population
Adults aged 22-64 years with migraine headache (based upon ICD‐9-code of ‘346’, ICD-10-code of ‘G43’) who were alive during the calendar years of 2012, 2014, and 2016, were included in the study.
Measures
Dependent Variables
Type and total healthcare expenditures
The types of healthcare expenditures, such as inpatient, outpatient, prescription, emergency room, and other healthcare expenditures (e.g., dental, vision, and durable medical equipment use, and others), were included in the analysis. Furthermore, the total healthcare expenditures, which consisted of the sum of all types of healthcare expenditures, were estimated. All healthcare expenditures were adjusted using the consumer price index and were expressed in 2016 constant dollars as provided by the US Bureau of Labor Statistics [21].
Key Independent Variable
The primary independent variable was the presentation of migraine, which included four mutually exclusive groups (migraine alone, migraine and anxiety, migraine and depression, migraine with both conditions).
Other Independent Variables
The other independent variables that were considered were the sociodemographic characteristics, which included gender, age in years (22-39, 40-49, and 50-64), race/ethnicity (White, African American, Latino and other), marital status (married, widowed, separated/divorced, and never married), education (less than high school, high school, and above high school), region of residence (Northeast, Midwest, South, and West), employment (unemployed, employed), health insurance (public, private, and uninsured), prescription medication insurance (insured, uninsured), and poverty status. Other independent variables included personal health practices such as physical activity (3 times/week, no exercise), smoking (smoker, non-smoker), perceived physical health (excellent/very good, good, and fair/ poor) and comorbid chronic health conditions such as diabetes, hypertension, and hyperlipidemia.
Statistical analysis
Descriptive statistics were used to describe the study sample across the four migraine groups. One-way analysis of variance (ANOVA) was used to compare the unadjusted means of the total healthcare expenditures across the four migraine groups. Generalized linear model (GLM) regressions with log link were used to estimate the healthcare expenditures associated with comorbid anxiety and depression among individuals with migraine after adjusting for the confounder (sociodemographic characteristics, health insurance status, and personal health practices, perceived physical health, and comorbid chronic health conditions). The GLM is an attractive alternative to OLS regressions on log-transformed expenditures because it corrects for heteroscedasticity and avoids the retransformation bias [22]. We used GLM with log link and gamma family distribution to estimate the adjusted medical expenditures associated with comorbid anxiety and depression. We also used a two-part model to estimate inpatient expenditures as the majority of the adults in the sample had zero inpatient expenditures. The first-part of the model estimates the probability of having zero expenditure versus positive expenditures. The second part of the model uses GLM to estimate the expenditures conditional on having positive inpatient expenditures. A P value of < 0.05 was considered statistically significant. Primary sampling unit, strata, and weights provided in the MEPS were all used in the analysis. All analyses were performed using survey procedures in the Statistical Analysis System SAS 9.4 (SAS Institute Inc., Cary, NC, USA) and the Software for Statistics and Data Science (STATA 15.1).