Inconsistent Medicaid Coverage is Associated with Negative Health Events for People with Epilepsy

Abstract

Objective.

Examine the association between gaps in Medicaid coverage and negative health events (NHEs) for people with epilepsy (PWE).

Methods.

Using five years of Medicaid claims for PWE, we identified gaps in Medicaid coverage. We used logistic regression to evaluate the association between a gap in coverage and being in the top quartile of NHEs and factors associated with having a gap. These models adjusted for: demographics, residence, medication adherence, disease severity, and comorbidities.

Results.

Of 186,616 PWE, 21.7% had a gap in coverage. The odds of being in the top quartile of NHEs per year were 66% higher among those with a gap (OR: 1.66; 95% CI: 1.61, 1.70). Being female, younger, and having psychiatric comorbidities increased the odds of having a gap.

Conclusions.

Gaps in Medicaid coverage are associated with being a high utilizer during covered periods. Specific groups could be targeted with interventions to reduce churning.

Introduction

Medicaid—a state-administered health insurance program for some low-income people, pregnant women, people with disabilities, some families and children, and some elderly— covers over 70 million individuals in the United States each year.^1,2 While the program is federally coordinated, eligibility/enrollment criteria and processes vary widely because it is administered by the states. Although policies vary across states, an individual often must renew their Medicaid every 12 months, but changes in income or employment or not having the support to help with renewal can result in disruptions in coverage.^3,4 Given these complexities, there has long been attention to gaps in Medicaid coverage, often referred to as churning when an individual moves in and out of coverage.^5–10 The attention to churning has only grown with the implementation of the Affordable Care Act and Medicaid expansion, with evidence suggesting that churning decreased after expansion.^3,10

Studies that explore the relationship between gaps and length in coverage and health outcomes have identified a few themes. People who have Medicaid gaps often have delays in preventive screening, higher rates of hospitalization for ambulatory care sensitive conditions, and high expenditures for those with depression or diabetes.^6,11–14 Similar trends also have been seen in analyses of private insurance and multi-payer datasets.^15,16 However, at least one other study did not find this same relationship, i.e., where gaps are associated with increased use of acute care.¹⁷ Many of these studies focus on a single state, a single facility, or a narrow time-frame (1–2 years) where the full effect of gaps or churning may not be apparent.

Epilepsy is a neurological condition that is characterized by unprovoked seizures and affects an estimated three million adults and nearly 500,000 children in the United States alone; it is a condition well poised to illuminate the impact of gaps in coverage on health outcomes.¹⁸ For people with epilepsy (PWE) lapses in coverage of prescriptions or follow-up care could mean experiencing a breakthrough seizure—risking physical injury or death, and importantly decreasing the quality of life for that patient. Numerous chronic physical and psychiatric conditions have been associated with epilepsy including somatic disorders, neurological disorders, mental health conditions, cognitive disorders, disabilities, and injuries.^19–30 Some of these conditions have a bi-directional association with epilepsy, further detailing the complexity of management and treatment for PWE. Further, PWE experience a number of social challenges including challenges in school, social relationships, employment, transportation (losing driving privileges due to recent seizures), independent living, and stigma.^{19,28,29,31–34} Taken together, these considerations make it clear that epilepsy provides a salient condition through which to understand the impact of gaps in coverage on health outcomes, and who is most at risk for those gaps.

Therefore, in this study we sought to explore the impact of gaps in Medicaid coverage on negative health events (NHEs), including hospitalizations and emergency department (ED) visits, for PWE, and subsequently to identify factors associated with gaps to inform potential interventions. This study overcomes a number of limitations of previous studies by using a substantially larger and multi-state dataset spanning a five-year period.

Methods

Data source.

This study used Medicaid claims data for five years (2010 – 2014) from 16 states that were available at the time this study was initiated: California, Georgia, Iowa, Louisiana, Michigan, Minnesota, Missouri, Mississippi, New Jersey, Pennsylvania, South Dakota, Tennessee, Utah, Vermont, West Virginia, and Wyoming. Medicaid expansion adoption status was obtained from the Kaiser Family Foundation at the time of manuscript preparation although expansion occurred only at the end of the study period.³⁵ While we could not examine the impact of Medicaid expansion, it is plausible that early adoption of Medicaid expansion is indicative of other policy-level factors conducive to facilitating continued enrollment that could have been in place pre-expansion; thus, this metric may help explain potential variation across states. We obtained county-level factors from the publicly available Area Health Resources File (AHRF) and matched them to the patient’s county of residence.

Inclusion criteria.

First, we identified people with epilepsy using previously published criteria.³⁶ Specifically, an individual was required to have a visit with an epilepsy or seizure diagnosis code (ICD-9-CM: 345.xx or 780.39); another visit for epilepsy or seizure 30 days later; and finally, at least two pharmacy claims for an anti-epileptic drug (AED) at least 30 days apart. We relied on a published list of AEDs for the third criteria.³⁶ We further limited our study population to those who had no missing geographic data and valid prescription fill data, those who were never dually eligible for Medicare during the study period, and those older than 18 at index date and younger than 65 by the end of the study.

Gaps in Medicaid coverage.

Using the monthly indicators for Medicaid coverage we were able to identify periods when an individual was not enrolled. We only considered the periods between known covered periods as gaps ignoring other gaps that occurred at the start or the end of the study period as these may reflect a point of entry/exit to/from the Medicaid program (Figure 1). That is, the data cuts for this study were artificial and we do not know if lack of enrollment at the start of the study was a gap from a time of previous enrollment or if they were newly enrolled. Similarly, we did not know if individuals who were not enrolled at the study’s end ever re-enrolled. We recorded whether or not an individual ever had a gap and the total number of gaps an individual had. These two variables were our independent variables of interest.

Figure 1.

Demonstration of how gaps in Medicaid coverage were counted in this study. As we did not know if non-covered periods at the beginning or end were true gaps, they we not counted as such.

Outcomes of interest.

The primary outcome of interest was negative health events (NHEs), which we defined as all-cause inpatient hospitalizations and emergency department visits. To account for varying amounts of observed time, and thus claims to capture these events, we annualized the number of NHEs to NHEs per year. To do this we took the number of NHEs, divided it by the individual’s total months enrolled, and multiplied by 12. For example, if someone had five NHEs in 53 months of enrolled time they would be said to have 1.13 NHEs per year. We then dichotomized this to those in the top quartile (the highest users) versus all others. While our primary outcome of interest was the top quartile of NHEs per year, we also disaggregated NHEs to examine those in the top quartile of hospitalizations per year and emergency department visits per year.

Covariates.

In order to adjust for factors potentially confounding the relationship between gaps in coverage and NHEs, we included the following as covariates: sex (male/female), race/ethnicity (White, Black, American Indian or Alaskan Native [AIAN], Asian or Pacific Islander [API], Hispanic, Native Hawaiian or Other Pacific Islander [NHOPI], and Other which included individuals with more than one race reported), age at index date, rurality (using the county-level Rural Urban Continuum Code)³⁷, state of residence, anti-epileptic drug (AED) medication adherence, if the individual was ever on a 3^rd unique AED (as a measure of disease severity), if the individual had undergone epilepsy surgery, whether an individual was ever in a nursing home,^38,39 the Elixhauser comorbidities,⁴⁰ and the density of primary care and neurologist physicians in the county as a binary factor, where 1 represents being in the bottom quartile of density. Epilepsy surgery and nursing home status were identified using Common Procedural Terminology (CPT)/Healthcare Common Procedure Coding System (HCPCS) codes (Supplemental Table 1, available from the authors upon request). We identified anti-epileptic drugs (AEDs), based on national drug codes from a published list of AEDs and calculated adherence using the proportion of days covered (PDC) where a PDC greater than 0.8 represented adherent.^41–43 We used all available claims to identify the Elixhauser comorbidities but required the condition to appear on at least one inpatient or two outpatient claims (30 days apart) to assign an individual as having that comorbidity. We grouped the Elixhauser comorbidities into psychiatric (alcohol abuse, drug abuse, depression, and psychoses) and physical conditions, and identified whether an individual had neither, physical, psychiatric, or both. Finally, we followed a similar process of annualizing NHEs to annualize and describe costs. We used the provided annual variable of the total amount Medicaid paid for the beneficiary in the year and report these annualized costs between those without and those with gaps.

Statistical analyses.

We created five multi-level logistic regression models to evaluate the association of a gap in coverage with being in the top quartile of NHEs per year (Model 1, unadjusted and adjusted) and the association of the number of gaps with being in the top quartile of NHEs per year (Model 2, unadjusted and adjusted). We then considered the two types of NHEs separately by evaluating the association between having any gap in coverage and being in the top quartile of hospitalizations (Model 3) or emergency department visits (Model 4) per year. The unadjusted models had no covariates, except random effect of state. Our fifth and final model sought to identify factors associated with having a gap in coverage. These models adjusted for demographics (race, sex, age), residence (rurality, primary care and neurologist density in the county of residence), medication adherence, disease severity (third AED, surgery), nursing home status, the Elixhauser comorbidities, and included random effects for the state of residence. To better examine the impact of the co-occurrence on physical and psychiatric conditions, we set having only physical comorbidities as the reference level. Finally, we replicated these analyses using the individual Elixhauser Comorbidities (Supplemental Tables 2 – 4, available from the authors upon request).

The data cleaning for this paper was generated using SAS software. Copyright © 2021 SAS Institute Inc. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc., Cary, NC, USA. Meanwhile, R version 3.6.3 was used for analysis.⁴⁴ This study was approved by the Institutional Review Board of Case Western Reserve University (Protocol #2018–0780), and the Privacy Board of the Centers for Medicare and Medicaid Services (CMS; Data Users Agreement #2019–52636).

Results

From an initial 301,014 individuals, 296,457 had no missing geographic data or valid prescription fill data, 273,864 were never dually eligible for Medicare during the study period, and 186,616 met the age criteria. Of these 186,616 adult people with epilepsy, 40,502 (21.7%) had at least one gap, and the median number of gaps was 1 (interquartile range: 1 – 2), while the mean was 1.4 (standard deviation: 0.77) (Table 1). Among those with gaps, compared with those without, there were fewer males (38.4% vs 46.0%), more White individuals (59.3% vs 52.8%), a similar proportion living in rural areas (21.7% vs 19.2%), and more younger adults (Table 1). There were clear differences across states: California, Georgia, Louisiana, Mississippi, New Jersey, and Tennessee had smaller percentages of PWE with gaps in coverage than the other states. Iowa, Michigan, Minnesota, Missouri, Pennsylvania, South Dakota, Utah, Vermont, and West Virginia had a larger percentage of people with gaps. California had the largest difference, representing 30.7% of those without gaps and only 22.9% of those with gaps. These trends did not seem to follow a discernable pattern with states that were early adopters of Medicaid expansion (Table 2).

Table 1.

Demographics of the study population

Demographics n (%) or median [IQR]	No Gaps in Coverage n = 146,114 (78.3%)	Gaps in Coverage n = 40,502 (21.7%)
Sex n (%)
Male	67,204 (46.0)	15,558 (38.4)
Race n (%)
White	77,216 (52.8)	24,012 (59.3)
Black	34,022 (23.3)	8,226 (20.3)
American Indian or Alaskan Native	1,007 (0.7)	488 (1.2)
Asian or Pacific Islander	2,049 (1.4)	404 (1.0)
Hispanic	15,648 (10.7)	4,532 (11.2)
Native Hawaiian or Other Pacific Islander	1,868 (1.3)	247 (0.6)
Other	14,304 (9.8)	2,593 (6.4)
Rurality
Rural	28,059 (19.2)	8,778 (21.7)
Age
18 to 24	25,514 (17.5)	9,290 (22.9)
25 to 34	35,046 (24.0)	12,265 (30.3)
35 to 44	30,023 (20.5)	9,334 (23.0)
45 to 54	36,091 (24.7)	7,214 (17.8)
55+	19,440 (13.3)	2,399 (5.9)
State
California	44,805 (30.7)	9,435 (23.3)
Georgia	11,947 (8.2)	2,474 (6.1)
Iowa	2,997 (2.1)	1,008 (2.5)
Louisiana	8,571 (5.9)	1,371 (3.4)
Michigan	14,634 (10.0)	6,016 (14.9)
Minnesota	6,310 (4.3)	2,980 (7.4)
Missouri	7,638 (5.2)	3,293 (8.1)
Mississippi	5,624 (3.8)	1,173 (2.9)
New Jersey	8,746 (6.0)	2,087 (5.2)
Pennsylvania	14,938 (10.2)	5,099 (12.6)
South Dakota	633 (0.4)	185 (0.5)
Tennessee	1,1856 (8.1)	2,443 (6.0)
Utah	1,973 (1.4)	689 (1.7)
Vermont	706 (0.5)	605 (1.5)
West Virginia	4,304 (2.9)	1,530 (3.8)
Wyoming	432 (0.3)	114 (0.3)
Gaps
Number of Gaps, Median [IQR]		1.00 [1.00, 2.00]
Number of Gaps, Mean (Std Dev)		1.41 (0.77)
NHEs, Hospitalizations, ED Visits
NHEs per year	1.60 [0.45, 4.15]	2.79 [1.15, 5.81]
NHEs per year (top quartile)	33,351 (22.8)	13,352 (33.0)
Hospitalizations per year	0.20 [0.00, 0.60]	0.27 [0.00, 0.81]
Hospitalizations per year (top quartile)	35,276 (24.1)	11,993 (29.6)
ED visits per year	1.33 [0.40, 3.40]	2.31 [0.92, 4.98]
ED visits per year top quartile	33,086 (22.6)	13,647 (33.7)
Costs
Costs per year, Median [IQR]	$16,276.80 [$7,947.39, $37,326.14]	$8,399.27 [$4,615.91, $15,604.52]
Costs per year, Mean (Std Dev)	$32,467.31 ($44,032.75)	$15,505.25 ($26,403.30)
Adherence
Adherent	74,169 (50.8)	13,462 (33.2)
Third AED
No Third AED	78,619 (53.8)	23,005 (56.8)
Surgery
Surgery	4,722 (3.2)	697 (1.7)
Nursing Home Status
Nursing Home	17,179 (11.8)	2,414 (6.0)
Comorbidities
None	27,510 (18.8)	8,999 (22.2)
Physical only	52,678 (36.1)	9,426 (23.3)
Psychiatric only	11,245 (7.7)	5,690 (14.0)
Physical or psychiatric	54,681 (37.4)	16,387 (40.5)
Area Level Measures
Primary Care Density (per 100,000)	69.35 [52.10, 85.82]	67.29 [52.01, 85.99]
Neurologist Density (per 100,000)	3.54 [1.54, 5.41]	3.04 [1.23, 5.41]

NHEs: Negative health events (emergency department visits + hospitalizations); ED: emergency department; AED: anti-epileptic drug; HIV/AIDS: human immunodeficiency virus/acquired immunodeficiency syndrome.

Table 2.

The number and percent of PWE with a gap in Medicaid coverage, stratified by state.

State	Medicaid Expansion Status*	n (%) with a gap
California, n = 54,240	Adopted (2014)	9,435 (17.4)
Iowa, n = 4,005	Adopted (2014)	1,008 (25.2)
Michigan, n = 20,650	Adopted (2014)	6,016 (29.1)
Minnesota, n = 9,290	Adopted (2014)	2,980 (32.1)
New Jersey, n = 10,833	Adopted (2014)	2,087 (19.3)
Vermont, n = 1,311	Adopted (2014)	605 (46.1)
West Virginia, n = 5,834	Adopted (2014)	1,530 (26.2)
Pennsylvania, n = 20,037	Adopted (2015)	5,099 (25.4)
Louisiana, n = 9,942	Adopted (2016)	1,371 (13.8)
Utah, n = 2,662	Adopted (2020)	689 (25.9)
Missouri, n = 10,931	Adopted, but not implemented	3,293 (30.1)
Georgia, n = 14,421	Not Adopted	2,474 (17.2)
Mississippi, n = 6,797	Not Adopted	1,173 (17.3)
South Dakota, n = 818	Not Adopted	185 (22.6)
Tennessee, n = 14,299	Not Adopted	2,443 (17.1)
Wyoming, n = 546	Not Adopted	114 (20.9)

^*As of February 2021

Among those with gaps compared with those without gaps, there was a higher percentage of individuals with no comorbidities (22.2% vs 18.8%), psychiatric comorbidities only (14.0% vs. 7.7%), and physical and psychiatric conditions (40.5% vs. 37.4%). Specifically, people with gaps had a higher prevalence of alcohol abuse (14.3% vs 9.8%), blood loss anemia (2.8 vs. 1.4%), depression (18.3% vs. 13.0%), drug abuse (24.3% vs 13.8%), and psychoses (40.0% vs 35.2%) (Supplemental Table 1, available from the authors upon request).

People with gaps had a higher number of NHEs per year with a median of 2.79 [IQR: 1.15, 5.81], compared with a median of 1.60 [IQR: 0.45, 4.15] for those without gaps (Table 1). This was true for the two components of NHEs, hospitalizations and emergency department visits, with a median of 0.27 [IQR: 0.00, 0.81] hospitalizations per year for those with gaps compared with 0.20 [IQR: 0.00, 0.60] (Table 1) for those without. For emergency department visits this difference was starker with 2.31 [IQR: 0.92. 4.98] visits per year for those with gaps compared to just 1.33 [IQR: 0.40, 3.40] for those without gaps (Table 1).

After adjusting for the covariates, the odds of being in the top quartile of NHEs per year were 66% higher among those with a gap in coverage (aOR: 1.66; 95% CI: 1.62, 1.70) (Table 3, Model 1). When evaluating the number of gaps, as a count variable, we see that each individual gap resulted in a 35% increase in the odds of being in the top quartile of NHEs per year (aOR 1.35, 95% CI: 1.33, 1.37) (Model 2, Table 3). Consistent with Model 1, after adjusting for covariates those with a gap had 60% higher odds of being in the top quartile of hospitalizations per year (aOR: 1.60; 95% CI: 1.56, 1.65), and 63% higher odds of being in the top quartile of emergency department visits per year (aOR: 1.63; 95% CI: 1.59, 1.68; Models 3 and 4, Table 3).

Table 3.

Models 1, 2, 3, and 4 adjusted odds ratios and 95% confidence intervals. Outcomes of interest include being in the top quartile of NHEs (Models 1 and 2), hospitalizations (Model 3), and emergency department visits (Model 4).

	Odds Ratio (95% CI)
Model 1: Association between Any Gap and NHEs
Unadjusted	1.66 (1.62, 1.70)
Adjusted	1.66 (1.61, 1.70)
Model 2: Association between the Number of Gap and NHEs
Unadjusted	1.35 (1.33, 1.37)
Adjusted	1.34 (1.32, 1.36)
Model 3: Association between Any Gap and Hospitalizations
Unadjusted	1.31 (1.28, 1.35)
Adjusted	1.60 (1.56, 1.65)
Model 4: Association between Any Gap and ED Visits
Unadjusted	1.74 (1.70, 1.78)
Adjusted	1.63 (1.59, 1.68)

NHEs: Negative health events (emergency department visits + hospitalizations)

In Model 5, we observed that being female, American Indian or Alaskan Native, Hispanic, 18 – 24 years old, non-adherent to AEDs, on a third AED, not having evidence of undergoing epilepsy surgery, not residing in a nursing home, having no comorbidities, only psychiatric comorbidities, or psychiatric and physical comorbidities, as well as living in an area with a low density of neurologists all increased the odds of having a gap in Medicaid coverage (Table 4). Notably, when compared to those who only had a physical comorbidity, those who had co-occurring physical and psychiatric comorbidities had a 61% increase in the odds of having a gap (OR: 1.61; 95% CI: 1.56, 1.66), and those with an isolated a psychiatric comorbidity had a 124% increase in the odds of having a gap (OR: 2.24; 95% CI: 2.15, 2.33). The impact of individual comorbidities was consistent with the pattern of the grouped comorbidities (Supplemental Tables 2 – 4, available from the authors upon request).

Table 4:

Adjusted odds ratios and 95% confidence intervals for Model 5. Outcome of interest is ever having a gap in Medicaid coverage.

Term	Adjusted Odds Ratio (95% Confidence Interval)
Sex
Male (ref: Female)	0.76 (0.74, 0.78)
Race
Black (ref: White)	0.89 (0.86, 0.92)
American Indian or Alaskan Native (ref: White)	1.45 (1.29, 1.63)
Asian or Pacific Islander (ref: White)	0.76 (0.68, 0.85)
Hispanic (ref: White)	1.21 (1.16, 1.26)
Native Hawaiian or Other Pacific Islander (ref: White)	0.64 (0.56, 0.74)
Other (ref: White)	0.72 (0.69, 0.76)
Rurality
Rural (ref: Urban)	1.00 (0.96, 1.03)
Age
25 to 34 (ref: 18 – 24)	0.91 (0.88, 0.94)
35 to 44 (ref: 18 – 24)	0.83 (0.80, 0.86)
45 to 54 (ref: 18 – 24)	0.58 (0.55, 0.60)
55+ (ref: 18 – 24)	0.40 (0.38, 0.42)
Adherence
Adherent (ref: Non-Adherent)	0.49 (0.48, 0.51)
Third AED
No Third AED (ref: Third AED)	1.30 (1.27, 1.33)
Surgery
Surgery	0.53 (0.49, 0.58)
Nursing Home Status
Nursing Home	0.68 (0.65, 0.72)
Comorbidities
None (Ref: Physical only)	1.54 (1.49, 1.59)
Psychiatric only (Ref: Physical only)	2.24 (2.15, 2.33)
Physical and psychiatric (Ref: Physical only)	1.61 (1.56, 1.66)
Area Level Measures
Primary Care Density: Bottom Quartile	1.06 (1.02, 1.09)
Neurologist Density: Bottom Quartile	1.00 (0.96, 1.04)

Discussion

This study of a vulnerable group of adult people with epilepsy, using a large, multi-year, multi-state dataset, found that gaps in Medicaid coverage are associated with more negative health events, as measured by hospitalizations and emergency department visits. This overall finding is consistent with research in other populations demonstrating an association between gaps in Medicaid coverage and utilization.^6,12–14,17 This study further clarified that this relationship is seen in both hospitalizations and emergency department visits, and it demonstrated an association between the number of gaps and NHEs. Finally, this work provided insight into important characteristics associated with having gaps. As previously discussed, epilepsy is a particularly salient condition for examining Medicaid churning because of its prevalence (over three million adults in the United States), complexity in disease presentation and management, high rates of comorbid conditions, and substantial social impact on individuals.^18,19,23,29 By itself, epilepsy would not qualify an individual for Medicaid coverage, however, those with more severe epilepsy that raises it to the definition of a disability per the Social Security Administration may qualify for Medicaid in their state without additional eligibility or needs.⁴⁵ In total, this work more clearly demonstrates the relationship between gaps in coverage and care utilization, and specific factors associated with gaps, going beyond just describing the prevalence of gaps in coverage or those who lose coverage.

We found that even after adjusting for demographics, residence, medication adherence, disease severity, nursing home status, and the Elixhauser comorbidities, PWE who had gaps in Medicaid coverage had increased odds for being in the top quartile of NHEs. Puzzling, however, was the difference in cost, adjusted for observed time, between those with gaps and those without gaps (Table 1). In our study we saw that people with gaps had lower odds of having severe disease (assessed via having a 3^rd unique anti-epileptic drug [AED] and undergoing surgery), so it is possible that more severe cases of epilepsy had greater motivation to stay enrolled in Medicaid, contributing to the higher costs observed. It is also possible that severe epilepsy lowers the probability of becoming transiently ineligible (e.g., through employment) or increases the probability of consistent re-enrollment by stimulating closer clinical observation (e.g., living in a nursing home). In this study we did not examine the specific reason for hospitalization or emergency department use, but future work should identify the causes of these NHEs to better identify what aspects of health may be most vulnerable to gaps in coverage.

Recognizing the importance of the association between gaps in coverage and NHEs, there have been a number of state-level interventions, many focused on children, to reduce gaps including reducing the frequency of re-enrollment, decreasing premiums, and extending eligibility.^46–49 Additional work has highlighted that often a misunderstanding of eligibility and general unfamiliarity with Medicaid are associated with barriers to enrollment and gaps.^50–52 In this study, we identified individual-level characteristics that could be areas of targeted intervention at the practice and policy levels. Based on our findings (Model 5), future interventions should be targeted to younger adults, those with less severe epilepsy (but who are enrolled in Medicaid), and those with psychiatric conditions. There was a clear dose-response relationship where older adults had significantly lower odds of having a gap than younger adults. One potential explanation for this is that social factors drive these gaps, and older adults may have acquired more experience navigating the Medicaid re-enrollment system over their life or have greater support or caregiving to help them navigate this system. Similarly, there is the potential for selection bias and competing risks, such that those with worse health and inconsistent coverage may face premature mortality. Additionally, we observed that psychiatric comorbidities increased the odds of having a gap in Medicaid coverage, with a substantially larger odds ratio among those without physical comorbidities. This indicates that patients with psychiatric conditions may be the most at risk for churning. Finally, this study identified potential racial and ethnic disparities, perhaps pointing to messaging or tailoring the communication of programmatic details, beyond the time spent during individual medical visits; this warrants greater attention and exploration.⁵⁰ With movement towards value-based payment, there will be growing incentives for health care systems to support their patients in maintaining Medicaid coverage, and the relationship between gaps and NHEs indicates that providing this support may reduce emergency department and/or hospital utilization—benefitting both the patient and the health care system. Policy makers should continue to explore opportunities, at the Medicaid policy and programmatic level, to assist and support health systems, practices, and their patients in enrolling and maintaining enrollment in Medicaid. For example, given our findings, policies could target additional support, or decrease the recertification frequency or burden, for patients with psychiatric comorbidities or those who live in certain geographic areas. Ideally, though, policy approaches would more generally reduce the burden of enrollment and re-enrollment in Medicaid to support all patients.

This study also underscores another critical component of Medicaid claims-based research: the role of gaps in study population and inclusion criteria. Traditionally, studies using Medicaid data limited their study population to those with continuous enrollment during the study period.^53,54 This is done to ensure claims completeness during the study period, a critically important point when conducting studies with administrative data. However, this study highlights that this exclusion step, particularly for studies that use multiple years of data, likely introduces selection bias to the work. Specifically, the issue is that patients who are high utilizers, with psychiatric and/or substance abuse disorders, and of specific race/ethnicities may be excluded and bias final estimates in studies of outcomes or disparities.

Despite this study’s strengths in using Medicaid claims data from a wide timeframe across 16 diverse states, it has limitations. First and foremost, the Affordable Care Act and associated Medicaid expansion have likely influenced gaps in Medicaid coverage, but the data used in this study, the most recently available at time of study initiation, were all from before, or just after, this expansion. At the start of the study period (2010), three of these states (CA, MN, NJ) were early adopters of Medicaid expansion.⁵⁵ While we did not observe a clear pattern between expansion and non-expansion states, future work should explicitly address the impact of Medicaid expansion on gaps in coverage. Thus far, work has largely shown that the Affordable Care Act may have decreased churning, and at least has not increased it - a large concern when Medicaid expansion was being implemented.^3,56 Work should continue to examine how the nature of expansion, and other changes to Medicaid such as work requirements, affect churning and gaps in coverage. Second, while this study used data from 16 diverse states, and observed notable differences in gaps by state, we were unable to assess specific reasons for these state-by-state differences. Future work should include a policy-level analysis to identify specific factors for these state-level differences that could serve as points of intervention. We observed interesting variation across states that, irrespective of expansion, warrant attention to understand whether there are specific policy-level differences that influence gaps in coverage.

An additional limitation is that we chose not to include non-covered months at the start or end of our study period as a gap as these times might have represented an entry to or permanent exit from Medicaid, and because the cut-off years of the study were not related to Medicaid enrollment. However, in sensitivity analyses we observed similar conclusions from models restricted to those people who were enrolled at the start and end of the study and whose only gaps, if any, occurred in the middle. Finally, while we identified factors associated with gaps, we were unable to assess the reason for the gap or re-enrollment. It is important to consider that for some individuals a gap in coverage may be due to improved circumstances, while for others it may be due to a lack of social capital or support. Individual-level interventions would best be developed if these root causes were known, as interventions might vary based on a wide variety of factors. For example, gaps due to income changes would rely more on federal and state policies to prevent them, whereas gaps due to missed or errors in re-enrollment could be addressed at the practice level. It is also possible that barriers to enrolling in Medicaid are non-uniform and thus it is the most supported individuals in some groups who are able to cross those initial barriers to enrollment. The larger issue of lack of attention to vulnerable, minority, and disadvantaged populations represents a societal problem and future work should continue to explore these complex relationships.

In conclusion, this study demonstrated the relationship between gaps in Medicaid coverage and being in the top quartile of hospitalizations and emergency department visits for people with epilepsy, and identified risk factors for churning. Those individuals with gaps had increased utilization while enrolled on Medicaid, even after controlling for a myriad of demographic and clinical factors. Future work should continue to explore the connection between gaps and negative health outcomes, including the number of gaps an individual has, and continue to identify and refine the the populations most at risk for having a gap in coverage.

List of Abbreviations

PWE

People with epilepsy

NHE

Negative health events

ED

Emergency department

AHRF

Area Health Resources File

AED

Anti-epileptic drug

CPT

Common Procedure Terminology

HCPCS

Healthcare Common Procedure Coding System

PDC

Proportion of days covered