MEPS is a set of surveys that collects data on health services that people use, their costs, payers, health insurance status of participants, and related data to generate nationally representative estimates of health care costs (the sum of direct payments for health care provided during the year)²² for the civilian, noninstitutionalized population.^23–27 We used the MEPS Household Component and Condition files. The former is drawn from households that participated in the previous year’s National Health Interview Survey.²⁷ Households are surveyed over five rounds over a 2-year period, and information is collected from about 13 000 households and 30 000 individuals annually. Asian, Black, and Hispanic individuals are oversampled. Data are collected via computer-assisted personal interviewing. The response rate to the full-year file, the main source of information about health care spending, was about 44% in 2017.²⁸

We identified respondents with records in the MEPS condition files for epilepsy or seizure. Respondents were asked to self-report conditions that bothered them and were associated with a health care encounter, prescription drug, or a disability day. More than 95% of the condition records for epilepsy and seizure were associated with a medical encounter or a prescription drug. Respondents’ descriptions of their conditions were recorded verbatim during the household interview and then translated to International Classification of Diseases codes at the three-digit level by professional coders. Interviewers prompt respondents to provide specific diagnoses rather than symptoms. Of MEPS respondents with an epilepsy diagnosis recorded in their medical record, 78.9% have a condition record associated with a medical event for epilepsy (operationalized in MEPS as “treated prevalence”).^19,20

To examine whether the MEPS condition records miss patients with epilepsy when a single definition is adopted,²⁰ we examined the proportion of respondents receiving antiseizure medications generally (e.g., prescribed for pain or anxiety, like benzodiazepines or gabapentin) and medications prescribed primarily for an epilepsy diagnosis (e.g., levetiracetam, lacosamide).²¹

Health care represented costs for physician office, hospital, emergency department, outpatient, home health, and dental services and prescription drugs and medical equipment. Costs represented payments from insurers to providers and respondents’ out-of-pocket costs. MEPS obtains payment amounts directly from respondents’ providers and pharmacies. We updated health care costs to 2019 US dollars using the Medicare Economic Index. We winsorized each cost category at the 99th percentile to downweight outliers.

We used the MEPS condition files to identify respondents with comorbid conditions (e.g., emphysema, diabetes) and known risk factors (e.g., traumatic brain injury, cerebrovascular accident) that would have an expected association with higher health care spending. We measured insurance status using the “INSCOV” variable, which indicates whether the respondent had any private or public coverage during the prior year. Otherwise, respondents were classified as being uninsured the entire year. Inclusion of insurance status and type in our study is important given the associations between socioeconomic factors, health care use, and shifting insurance status among epilepsy and seizure patients.^8,29

Records for respondents included in the MEPS for 2 consecutive years were treated as independent observations. The MEPS stratum and primary sampling unit variables account for clustering at the respondent level. We matched each respondent with a condition record for seizures, epilepsies, or seizure or epilepsies to 10 control respondents based on age and sex. We performed all analyses in Stata version 16.0 (StataCorp). All analyses were weighted for the MEPS complex survey design.

We compared spending between cases and controls using generalized linear models with a Gamma distribution and a log link. For each condition and cost category (e.g., overall, inpatient), we estimated three models. The first included only a constant term, the second adjusted for patient demographic characteristics (age group, sex, race/ethnicity, marital status, region) and education level, and the third further adjusted for comorbidities (Table S2).

We omitted insurance status as a covariate, because it was not independent. Seizures or epilepsies may cause persons to lose employment and lose access to employer-sponsored insurance. Conversely, people with seizures or epilepsy may be more likely to buy individual coverage or enroll in government programs to obtain care.⁸

Comparing estimates across models allowed us to assess the sensitivity of estimated cost differences to patient characteristics. We reported marginal effects (or “predicted marginals”), which represented the expected difference in spending following adjustment. Using the full model, we also compared total expenditures between cases and controls by subgroups. We combined estimates of the prevalence of persons with seizures, epilepsies, or seizures or epilepsies with estimates of the average and incremental costs per case and the total population covered by MEPS to calculate the aggregate annual cost of each condition.

We estimated annual growth rates by estimating two-part models separately for cases and controls that included a time trend variable. We divided the marginal effect of the time trend variable by average spending across the entire period to calculate growth rates. We evaluated the significance of differences in growth rates by estimating two-part models on the entire sample (cases and controls) that included separate time trend variables for cases and controls. We used a Wald test to examine the significance of these differences in the time trend. We estimated total spending by multiplying estimates of disease prevalence by regression-adjusted estimates of spending attributable to epilepsy or seizure.

Researchers and users with approved research projects can access MEPS data files that have not been publicly released for reasons of confidentiality.^19,20

We identified 2344 persons with a condition record for seizure, and 1078 persons with condition records for epilepsy. Few respondents, 89 for seizure and 53 for epilepsy, had a condition record that was not associated with a medical event. A total of 3422 respondents had either a record for epilepsy or seizure.

Among respondents with seizures (no epilepsy code), 1359 (58.0%) had a prescription for antiseizure medication and 867 (37.0%) had a prescription for a drug used in patients with epilepsy (e.g., levetiracetam). Among respondents with epilepsy, 729 (67.6%) had a prescription for antiseizure medication and 558 (51.8%) had a prescription for a drug used in patients with epilepsy (Table S3). Among the 290 791 respondents without condition records for epilepsy or seizure, 19 557 (6.8%) received an antiseizure medication and 2187 (.8%) had a prescription for a drug used in patients with epilepsy.

Seizure prevalence was .76 (95% confidence interval [CI] = .71–.81), and epilepsy prevalence was .38% (95% CI = .34–.41; Table S4). The treated prevalence of epilepsy or seizure was 1.14% (95% CI = 1.08–1.20).

After matching on age and sex, the distributions of cases and controls across age and sex groups were similar (Table 1). Remaining differences were likely due to sample weights to the matched sample. Epilepsy cases were more likely to be non-Hispanic White, to be married, and to have graduated from high school but not college.³⁰ Comorbidities were more common among epilepsy and seizure cases than controls (Table S2).

Compared to their controls, a similar proportion of respondents with epilepsy were uninsured for the entire year, 8.4% versus 10.8% (estimates reflect sample weights). Respondents with epilepsy were more likely to have public coverage (37.2% vs. 21.4%) but less likely to have private coverage (54.4% vs. 67.8%). The proportion of respondents with epilepsy who were uninsured declined from 12.8% in 2010 to 3.4% in 2018 (Table S5).

Average annual spending was $11 333 for persons with epilepsy, $16 951 for persons with seizures, and $15 096 for persons with epilepsy or seizures (Table 2). Regression-adjusted differences in spending between cases and controls declined as more covariates were added to the model. The unadjusted difference in spending between epilepsy or seizure cases and controls was $9385, and the regression-adjusted difference from the model that included all controls was $6835 (95% CI = $5623–$8084; Table S6). Adjusted differences from the full model were $4580 (95% CI = $3362–$5798) for epilepsy cases versus controls and $7935 (95% CI = $6237–$9634) for seizure cases versus controls (Table S6).

Spending in the “other” category, which includes outpatient care, accounted for roughly $3052 (95% CI = $2176–$3929) of the difference in overall spending between epilepsy or seizure cases versus controls, and inpatient care and prescription drugs accounted for equal amounts (approximately $1400; Table 2). Average out-of-pocket costs for epilepsy or seizure cases were $991, and the difference in out-of-pocket costs between cases and controls was $213.

Spending overall, and prescription drug and outpatient costs were significantly higher in 2012 and in subsequent years among epilepsy cases versus controls (Figure 1A–D). Annualized growth rates over the period 2010–2018 were higher for cases compared to controls (Table 2). For example, the growth rate in total spending incurred by epilepsy or seizure cases was 7.6% compared to 3.6% among controls. Differences in spending growth between cases and controls were especially large for inpatient care (15.1% vs. 5.0%) and prescription drugs (8.8% vs. 1.9%; Table 2). Out-of-pocket spending did not change much over the study period.

Multiplying our estimates of prevalence (first row of Table S5) by our estimates of the incremental cost per case from the regression model that includes all covariates (Model 3 in Table 2) and by the population covered by MEPS (314 million), we estimate that total costs for persons with epilepsy or seizure was $54.0 billion nationally (Figure 2). Incremental costs, which capture the added contribution of epilepsy or seizure, were $24.5 billion.

This study has several limitations contributing to an underestimation of costs.