Duration and Continuity of Medicaid Enrollment Before the COVID-19 Pandemic

Key Points Question How stable was the continuity of Medicaid enrollment from 2016 to before the COVID-19-related moratorium on disenrollment, and what factors contributed to stability? Findings This cross-sectional analysis of nationally representative Medicaid data on 5.7 million persons show that Medicaid expansion was associated with less churning (defined as one or more breaks in coverage) and longer duration of coverage, whereas the use of ex parte reviews was associated with less churning but shorter duration of coverage. Even after controlling for demographic and state policy factors, there were very large interstate differences in Medicaid continuity. Meaning After the federal COVID-19 public health emergency declaration expires and the disenrollment moratorium ends, states should consider adopting Medicaid policies that stabilize the continuity of coverage.

We found a small number of data problems and cleaned them where possible. For example, we found 2,876 cases out of the 5.7 million sampled cases where the person has zero months of Medicaid enrollment and we eliminated them. In a few cases, we found age data that appeared inconsistent with the ages that should have been possible for their eligibility category, e.g., children under 18 in the expansion adult category; we eliminated these as well. The biggest gap was that 19.5% of enrollees had "unknown" race. It is well known that reporting of race/ethnicity in Medicaid is often incomplete; we included those cases and made Unknown as a separate category of race/ethnicity. It is not clear from the documentation whether race/ethnicity was self-reported by enrollees or determined by staff; it is plausible that both methods are used.
Data about state policies for 2016 came from the Kaiser Family Foundation (KFF), which surveyed state Medicaid agencies about their enrollment policies. 3 Of particular importance, we used data about ex parte reviews in renewing Medicaid eligibility, which means that agencies look at other data available to them to conduct renewals. For example, if a person's Medicaid certification period expires in March, but SNAP eligibility was determined in January, the SNAP data can be used to renew their coverage and there is no need to ask for a renewal application. Federal regulations require the use of ex parte reviews for those whose incomes is based on Modified Adjusted Gross Income (MAGI) criteria, but the KFF data indicated that only 34 states reported using it in 2016. KFF also asked what percent of renewals were "completed via ex parte" (< 25%, 25%-50%, 50%-75%, and 75%+) but 8 of the 34 states did not have an estimated percentage. We were unsure about how to interpret the percentages for four reasons: (1) Not every renewal is a MAGI case (e.g., aged, disabled and foster enrollees are typically not MAGI) and it was not clear if these cases were in the denominator or not. (2) Medicaid enrollment is typically much larger than SNAP or TANF enrollment and we would not expect all (or even a majority) of Medicaid enrollees to have alternative data available. Some states may use wage matching data from unemployment systems, but not all do and wage-matching data are often incomplete and not current. (3) Does having a renewal completed via ex parte mean that the other information led to a renewal or that a check was completed, even if they found no other data and then asked the person to submit a renewal form? (4) In principle, given the federal regulation, all states are supposed to use ex parte reviews. Did the fact that 17 states did not report ex parte mean they are not doing it or that it was not reported to KFF? Given this uncertainty, we just used ex parte reviews as a binary variable of used ex parte or not. In contrast, a recent MACPAC report used the percentage of ex parte reviews using 2018 KFF data in which 46 states reported using ex parte reviews and found that greater use of ex parte was associated with longer continuity, although it did not test for statistical significance nor controlled for other demographic and policy factors. 4

Enrollee Demographic Characteristics
There were a limited set of data about enrollees. Table A-1 provides a tabulation of the percentage of enrollees in each of six Medicaid eligibility characteristics by age category, race/ethnicity and gender, in states that expanded Medicaid as of January 1, 2016 and those that did not. Expansion states had a larger percentage of enrollees who were adults 18 to 64 years old than non-expansion states and therefore a smaller percentage who were children 0 to 17. Expansion states had a higher percentage of enrollees who were Latino, White or Other, while non-expansion states had higher percentage of enrollees who were Black or Unknown race/ethnicity. Expansion and non-expansion states had similar gender patterns, although traditional adult enrollees were more likely to be female in both expansion and non-expansion states.

Medicaid Managed Care Organization Enrollment
Over two-thirds of Medicaid enrollees receive care from Medicaid managed care organizations (MCOs), which are responsible for the organization and delivery of health services for their members (although some Medicaid services or medications may be "carved out" and still under fee-for-service systems). The MCOs determine the networks of providers who are available for care for their members, although sometimes members may see out-of-network providers. Some Medicaid enrollees are not in MCOs because the state has not established Medicaid MCOs for some categories (e.g., the aged) or because some are participating in another form of managed care, Primary Care Case Management, which does not require MCO membership.
Table A-2 provides information about MCO membership and the average duration of MCO enrollment by eligibility category. We define MCO membership as having at least one month enrolled in an MCO. As can be seen, Medicaid children and expansion adults are the most likely to be enrolled in MCOs (93% and 88%, respectively), followed by CHIP children, traditional adults and people with disabilities (75%, 74% and 73%, respectively). Aged beneficiaries, almost all of whom are dually enrolled in Medicare, are least likely to be in Medicaid MCOs (50%).
The average months of MCO membership is displayed in two ways. First is the average number of months enrolled in MCOs, using all Medicaid enrollees as the denominator. By this measure, the average span in an MCO is consistently shorter than the average months of enrollment shown in Table 1, between 4.9 and 8.7 months enrolled. But when the average duration is computed only for those with at least one month of MCO membership, the average increases to between 8.4 and 10.0 months, still slightly lower than the overall average number of months enrolled, as seen in Table 1.
One reason the duration of enrollment in MCOs is shorter is that new Medicaid enrollees are not automatically enrolled in MCOs, but are usually given one or two months to select an MCO (if one is available for them). Thus, they first spend on two months in fee-for-service care prior to enrollment in an MCO. If they do not make a selection, they are auto-enrolled in an MCO selected by the state agency, although they typically have a grace period to switch plans if they wish. When they renew coverage, the state may automatically assign them to the same MCO they had, offer them an option to change plans or auto-assign them if no choice is made. The somewhat briefer time spent in an MCO affects the delivery and financing of care in Medicaid, since they spent less time covered by the MCO and participating in the MCO's provider network. MCOs incur some administrative costs on-boarding new enrollees, so this also affect MCO finances. 5

Multivariate Analyses with State Fixed Effects
As noted in the main text, our multivariate models of factors associated with the number of months enrolled in Medicaid and with having breaks in coverage incorporated state fixed effects, implemented by adding a binary dummy variable for every state except Texas. Adding the fixed effects helps control for other unmeasured state characteristics that might also affect Medicaid continuity. Texas was selected as the omitted reference state since initial analysis showed it had the lowest average duration among the states.
The coefficients for the state fixed effects are shown in Tables A-3 and A-4, which correspond with Tables 3 and 4 in the main text but add the state-level coefficients. Fixed effects coefficients could not be estimated for a few states. As seen in Tables A-3 and A-4, coefficients for Vermont, West Virginia and Wisconsin could not be estimated due to collinearity between the policy variables and the other state IDs and policy variables. In Table A-4, the model about breaks in coverage, three more states (Kentucky, New York and Wyoming) were not included because they each reported no breaks in coverage. Nonetheless, we could estimate coefficients for state fixed effects for most variables.
A noteworthy finding is that, even after controlling for demographic and state policy variables, there were large and usually highly significant coefficients or odds ratios for the state fixed effects. For example, in Table A-3, the coefficient for Missouri (4.1, 95% CI = 4.0 to 4.2) means that even after controlling for the demographic and policy variables, Missouri's average enrollment duration is about 4 months longer than that of Texas, the omitted reference state. This is very close to the unadjusted difference of 4 months between the Missouri and Texas in Table 2. But while Texas had the lowest unadjusted average number of months in Table 2, the ordinary least squares regression model in Table A-3 indicates that some states like Georgia or North Dakota have even lower duration than Texas, after controlling for other factors.
The odds ratios for the state fixed effects in Table A-4 indicate that most states have a lower odds of breaks in coverage than Texas, the omitted reference state, after controlling for the demographic and state policy variables.
The presence of large, unexplained differences in the average duration of Medicaid enrollment and the odds of coverage breaks across the states suggest that there could be other unmeasured demographic, policy or operational factors that contribute to large interstate differences in the continuity of enrollment in Medicaid. Our policy variables included state Medicaid expansions and ex parte reviews, although we also tested the association of telephone renewals and pre-populated renewal forms with continuity of coverage, but did not find significant associations. We could not control for immigration status, employment, health status, marital status or the presence of children in the family because those variables were not available the SyH-DR data. There may also be other policy factors, such as the length or complexity of the renewal process, the extent to which states had current correct addresses for their enrollees, or the availability of staff to help applicants complete renewals that might also affect the continuity of enrollment.