The data are from the household component of the Medical Expenditure Panel Survey (MEPS), a nationally representative survey conducted by the Agency for Healthcare Research and Quality. This analysis utilizes four overlapping 2-year cycles — 2000–2001, 2001–2002, 2002–2003, and 2003–2004. Data for each individual cover 24 months and include information on health insurance for each month, income, health status, and demographics.
The sample consists of all adults aged 18–63 (to exclude those becoming Medicare-eligible) who were enrolled in Medicaid at any point during the survey. Secondary analyses include children meeting the same criteria. To fully characterize Medicaid disenrollment, this paper presents two approaches.
The first approach assesses insurance coverage after 12 months for all individuals enrolled in Medicaid at the survey’s outset (3,141 adults
6,014 children). This analysis estimates overall annual disenrollment among all Medicaid enrollees — including those who had been enrolled for long periods of time.
The second approach is a survival analysis of all newly enrolled individuals (1,851 adults
2,545 children), which provides detailed results on the lengths of typical Medicaid enrollment periods, as well as “churning” — when disenrollees return to Medicaid at a later date.
Both approaches are necessary because the two groups represent distinct populations that provide different information. The first group — those enrolled at the survey’s outset — is representative of all people in Medicaid at any point in time. The second group — those who newly enroll during the study — is representative of all spells of enrollment in Medicaid over time.
In the first analysis, outcome measures were the survey-weighted percentages of individuals who, 12 months later, were still enrolled in Medicaid, acquired other health insurance, or became uninsured.
In the survival analysis, the outcome measure was the duration of continuous Medicaid coverage after initial enrollment. Kaplan-Meier curves were estimated for the length of coverage in months, with right-censoring of individuals whose initial enrollment continued until the survey’s completion. Cox tests for equality of survivor functions (survey-weighted univariate Cox proportional hazards models) were performed for comparisons between adults versus children, and among adults with varying eligibility-related characteristics discussed below.
Previous research categorized disenrollment into three categories: lost eligibility, acquiring new insurance, and “drop-out” — when eligible individuals lose Medicaid coverage and become uninsured.5
Thus, survey-weighted proportions were calculated to determine how many disenrollees had — by 6 months after leaving Medicaid — acquired other health insurance, reenrolled in Medicaid, or become uninsured. In terms of lost eligibility, Medicaid eligibility is set by each state within broad federal guidelines. The publicly available MEPS does not provide state-identifiers. Loss of eligibility typically occurs through two routes — loss of categorical eligibility (disability, welfare, or pregnancy) or increases in family income. To assess the effect of eligibility on disenrollment, despite the lack of state-identifiers, two additional analyses were conducted.
First, a survival analysis was conducted for a subset of adults — the “always eligible” — who satisfied federal minimum requirements for eligibility throughout their enrollment period: Supplemental Security Income (SSI) disability receipt, receipt of cash public assistance (Temporary Assistance to Needy Families, or TANF, commonly known as welfare), or pregnant women below 133% of the Federal Poverty Line (FPL).12
The MEPS does not assess pregnancy, so pregnancy was imputed for all mothers having children under age 1. Pregnancy-linked eligibility was imputed from 9 months prior to 2 months after the child’s birth, in accordance with Medicaid rules. If retention were low even among the “always eligible,” this would indicate that Medicaid disenrollment is not simply due to lost eligibility.
Second, a survival analysis was performed with the sample divided into those whose nominal income stayed the same or decreased, those experiencing a small increase ($0 to $9,999), and those with larger income gains (>$10,000). If most disenrollment is due to individuals becoming too wealthy to qualify for Medicaid, then this analysis should indicate significantly more disenrollment among those with increased income over time.
Cox proportional hazards models were then estimated, focusing on “drop-out” specifically: the outcome variable was Medicaid disenrollment without reacquiring health insurance within 6 months, and the regression controlled for eligibility measures. Eligibility variables were family income (in 2006 dollars), poverty status (% FPL), pregnancy, SSI-disability, and TANF participation.13
Health-related variables were self-reported health status and mental health, any health-related limitations at work, and the following diseases: hypertension, diabetes, asthma, emphysema, stroke, coronary artery disease, arthritis, and depression/anxiety. Demographic covariates were race/ethnicity, gender, age, education, whether the individual had children enrolled in Medicaid, Medicaid-managed care participation, family size, geographic region, and how many months into the survey the individual’s Medicaid enrollment began.
All data analyses were conducted using Stata 7.0 and accounted for the MEPS’ complex survey design.14,15