Medicaid Coverage ‘Cliff’ Increases Expenses And Decreases Care For Near-Poor Medicare Beneficiaries

Abstract

Cost sharing in traditional Medicare can consume a substantial portion of the income of beneficiaries without supplemental insurance from Medicaid, an employer, or a Medigap plan. Near-poor Medicare beneficiaries (with incomes over 100 percent but under 200 percent FPL) are ineligible for Medicaid but frequently lack alternative supplemental coverage, resulting in a supplemental coverage “cliff” of 25.8 percentage points just above the eligibility threshold for Medicaid (100 percent FPL). We estimated that beneficiaries affected by this supplemental coverage cliff incurred an additional $2,288 out-of-pocket spending over two years, used 55 percent fewer outpatient evaluation and management services per year, and filled fewer prescriptions. Lower prescription drug use was partly driven by low take-up of Part D subsidies, which Medicare beneficiaries automatically receive if they have Medicaid. Expanding eligibility for Medicaid supplemental coverage, and increasing take-up of Part D subsidies, would lessen cost-related barriers to health care use among near-poor Medicare beneficiaries.

The traditional Medicare program, which covers 40 million people, includes substantial beneficiary cost sharing.¹ Part A of traditional Medicare, which covers inpatient and skilled nursing facility care, has an inpatient hospital deductible ($1,484 per benefit period in 2021) and daily coinsurance of $186 for skilled nursing facility stays lasting 21–100 days (longer stays are not covered). Part B of traditional Medicare, which covers outpatient and professional care, includes an annual deductible ($203 in 2021) and 20% coinsurance for most services.²

Medicare beneficiaries can limit their exposure to these costs by obtaining supplemental insurance through Medicaid, an employer, or a private Medigap plan. Approximately 80% of people with traditional Medicare have supplemental coverage, but prominent coverage gaps exist.³ One of these gaps is among near-poor Medicare beneficiaries with incomes >100% to <200% of the federal poverty level (>$12,880 to <$25,760 for a single person in 2021).^1,4 These near-poor individuals, who account for nearly 30% of the Medicare population, exceed the income limit for Medicaid supplemental coverage but are less likely to have private supplemental insurance than those with higher incomes.^1,5

Recent estimates suggest that 40% of near-poor Medicare beneficiaries spend at least one-fifth of their income on health care costs.¹ High out-of-pocket costs raise concerns that low-income beneficiaries without supplemental coverage may forego care.³ These concerns are underscored by evidence that patients of lower socioeconomic status are more likely to reduce the use of outpatient care and prescription drugs subject to cost sharing, which may have adverse consequences for health.^6–9 Such effects of cost sharing are at odds with its intended purpose, which is to promote more efficient health care use by holding patients accountable for a portion of spending.^10,11 Consequently, some policy analysts have proposed expanding eligibility for Medicaid supplemental coverage to as high as 200% FPL.^12,13 However, no research has examined the effects of Medicare’s cost sharing in populations that would be targeted by these reforms.

We examined the consequences of cost sharing in traditional Medicare by studying a supplemental coverage “cliff” among near-poor beneficiaries. This cliff arises from the income eligibility threshold for Medicaid supplemental coverage at 100% FPL. Beneficiaries whose income slightly exceeds this threshold are less likely to have any supplemental coverage, and thus are more likely to face Medicare’s full cost sharing, than those just below it. We leverage this abrupt difference in exposure to cost sharing in a regression discontinuity design. This study design helps to isolate effects of the coverage cliff from confounders to the extent that Medicare beneficiaries above and below Medicaid’s eligibility threshold are similar.¹⁴ We examined effects of this coverage cliff on out-of-pocket spending and Part A and B utilization. Because Medicare beneficiaries with Medicaid automatically receive Part D subsidies, we also studied spillovers on prescription drug use among Part D enrollees.^15,16

METHODS

Medicaid supplemental coverage

Approximately 11 million Medicare beneficiaries receive Medicaid through several state-administered programs with different benefits and eligibility criteria. Beneficiaries with income ≤100% FPL and low assets are eligible for the Qualified Medicare Beneficiary (QMB) program, which covers all Part A and B cost sharing. (Services excluded from traditional Medicare, such as long-term nursing home care, are also covered if a QMB recipient meets state-specific eligibility criteria for full Medicaid; see Appendix Exhibit 1 for details¹⁷). Beneficiaries with income >100% to ≤135% FPL and low assets qualify for Medicaid assistance with the Part B premium, but not Part A or B cost sharing.¹⁸ Thus, Medicaid’s eligibility rules introduce a coverage cliff at 100% FPL, which we refer to as the income eligibility threshold for Medicaid supplemental coverage.

Study data and population

We analyzed the 2008–2016 biennial waves of the Health and Retirement Study (HRS), a longitudinal survey of the US population aged 50 and older,¹⁹ linked to Medicare enrollment and claims files from 2007–2016. These data enabled us to assess respondents’ eligibility for Medicaid, receipt of supplemental coverage, out-of-pocket spending, and use of care. Over 80% of respondents with Medicare provide their beneficiary identification number to enable linkage of Medicare data. The HRS oversamples Black and Hispanic individuals,²⁰ who disproportionately have low to moderate incomes, making this survey ideal for our study.

From the HRS, we selected respondents who lived in the community at the time of the survey, were linked to Medicare files, and continuously enrolled in traditional Medicare (Parts A and B) in the survey year and prior year (for decedents, we required continuous enrollment until death). We excluded respondents in Connecticut, the District of Columbia, and Maine (all study years) and Indiana (on or after 2014), as these states provided Medicaid supplemental coverage to Medicare beneficiaries with incomes >100% FPL.

We restricted our sample to respondents with income <200% FPL in a survey wave. We identified this sample, and respondents who qualified for Medicaid supplemental coverage, by applying Medicaid income counting rules to income data from the RAND HRS. Income data in the RAND HRS are imputed for respondents who did not report their exact income. To minimize error, imputations are bounded within income brackets reported by respondents.²¹ We excluded 103 respondents with incomes of 96–104% FPL to limit attenuation bias from error in reported income (Appendix¹⁷).

We linked each HRS wave to Medicare files from the survey year and preceding year, because the HRS asks respondents to report income for the year preceding each survey wave and we wanted to examine outcomes in all years for which we had linked Medicare data.

Study variables

Supplemental coverage

We used Medicare enrollment files to assess respondents’ receipt of Medicaid supplemental coverage through the QMB program (in each study year) and the HRS to assess self-reported receipt of private supplemental insurance from an employer or Medigap plan (in each survey year). We constructed a respondent-year-level indicator of having either Medicaid or private supplemental coverage, applying the HRS indicator of private coverage to both the survey year and prior year.

Study outcomes

Out-of-pocket costs:

We analyzed self-reported out-of-pocket medical spending over the prior two years from the RAND HRS. These data include imputed values for respondents who do not report exact spending.²¹ We analyzed out-of-pocket spending in aggregate and for physician office visits, hospital care, and other medical care (excluding prescription drugs, which we studied for Part D enrollees). Because one function of insurance is to mitigate large losses, we examined whether respondents incurred catastrophic spending, which we defined as 2-year out-of-pocket spending >$2,000. We chose this threshold because, for an individual at 100% FPL, such spending would exceed approximately one month of income.⁴ In sensitivity analyses reported in the Appendix,¹⁷ we examined thresholds of >$1,000 and >$3,000.

Part A and B utilization:

We measured the following outcomes from Medicare claims at the level of the respondent-year: outpatient services in physician offices or hospital outpatient departments (in aggregate and categorized by Berenson-Eggers Type of Service codes); emergency department visits and hospital observation stays; inpatient admissions; admissions for ambulatory care-sensitive admissions; and admissions to skilled nursing facilities.

Part D outcomes:

We examined Part D enrollment, and among Part D enrollees (66.3% of our sample), we analyzed the receipt of prescription drug subsidies, spending, and utilization. Although Medicaid does not subsidize prescription drugs, there are several mechanisms through which a Medicaid coverage cliff could have spillovers on medication use. First, Medicare beneficiaries automatically obtain the Low-Income Subsidy (LIS), which reduces Part D cost sharing, if they receive Medicaid supplemental coverage or Medicaid subsidies for the Part B premium.¹⁵ Prior work demonstrated that differences in the take-up of these Medicaid benefits leads to a cliff in LIS enrollment immediately above 100% FPL.¹⁶ Because the value of the LIS can be substantial for people needing multiple or high-cost medications,²² this LIS enrollment cliff could affect medication use. Second, by reducing out-of-pocket costs for medical care, Medicaid may increase beneficiaries’ ability to afford medications. Third, patients who receive fewer outpatient visits because they face higher Part B out-of-pocket costs may have fewer opportunities for physicians to prescribe medications.

Therefore, among Part D enrollees, we assessed annual out-of-pocket Part D spending, LIS payments for Part D-covered drugs, and medication use (measured by number of fills) among Part D enrollees. We also assessed annual fills in 22 classes of drugs used to treat prevalent chronic conditions in our sample (see Appendix for the drug classes¹⁷).

Respondent-level covariates

We used Medicare enrollment files to assess respondents’ age, sex, race/ethnicity, disability (original reason for Medicare entitlement), and presence of end-stage renal disease. We used the HRS to assess marital status, education, lifetime tobacco use, presence of dementia, and use of a helper to complete the HRS (a proxy for functional impairment). Finally, we included measures of health status from the prior HRS wave: difficulties performing activities of daily living and instrumental activities of daily living, and indicators of ever having been diagnosed with arthritis, cancer, diabetes, heart disease, hypertension, lung disease, a behavioral health disorder, or a stroke. (We included measures from the prior wave to avoid controlling for any contemporaneous health effects of Medicaid coverage.)

Statistical analyses

We used a regression discontinuity design to examine effects of the supplemental coverage cliff among near-poor Medicare beneficiaries. We conducted our analyses in two stages.

First, we ran a respondent-year-level linear regression model to estimate the difference, immediately above vs. below the Medicaid eligibility threshold, in the receipt of Medicaid or private supplemental coverage, which we adjusted for income, the covariates described above, and state and year fixed effects. Hereafter, we refer to estimates of differences immediately above vs. below the Medicaid threshold as discontinuities.

Second, we assessed discontinuities in study outcomes by estimating linear regression models of the form described above. We estimated these models at the level of the respondent-year for outcomes in Medicare data and respondent-survey wave for outcomes in the HRS.

We combined estimates from these first and second stage models using instrumental variable (IV) methods to estimate effects of the supplemental coverage cliff. For analyses of out-of-pocket spending and Part A and B utilization, we used an indicator for exceeding Medicaid’s income eligibility threshold as an IV for lacking supplemental coverage. The resulting IV estimates give the effect of the coverage cliff among Medicare beneficiaries who lacked supplemental coverage because their income just exceeded the Medicaid eligibility threshold.¹⁴ For brevity, we refer to these estimates as effects of the supplemental coverage cliff. These estimates equal the discontinuity in a study outcome (from the second stage model) divided by the absolute value of the discontinuity in supplemental coverage (from the first stage model).

In analyses of Part D outcomes, we estimated effects of the cliff in LIS enrollment above the Medicaid eligibility threshold. For these outcomes, we used an indicator that income exceeded the Medicaid threshold as an IV for lacking the LIS. We refer to the resulting IV estimates as effects of the LIS enrollment cliff. Additional details of our analyses are in the Appendix.¹⁷

To assess the extent to which we could isolate effects of the coverage cliff from confounders, we examined whether respondent characteristics trended continuously through Medicaid’s eligibility threshold. We also ran analyses stratified by disability status. We did this because of institutional linkages among programs that facilitate entry into Medicaid and Medicare for low-income people with disabilities (Supplemental Security Income and Social Security Disability Income), which result in a higher proportion of low-income individuals qualifying for Medicare through a disability pathway.²³

We adjusted for survey weights to produce estimates that are nationally representative of the community-dwelling Medicare population age 50 and older with income <200% FPL. We used robust variance estimation to account for clustering within households.

Supplementary analyses

First, we assessed whether respondents might have manipulated their income to attain Medicaid eligibility, as this would bias our study design. Specifically, we examined the distribution of respondents around the Medicaid eligibility threshold, assessed income changes across the threshold, and compared the characteristics of respondents with income changes that resulted in a change in Medicaid eligibility. Second, we examined discontinuities in the receipt of other means-tested assistance (e.g., the Supplemental Nutrition Assistance Program). Such discontinuities do not invalidate our study design, since only Medicaid subsidizes health care costs, but they do highlight that people may differ in their eligibility for Medicaid and other programs that augment the resources of low-income households. Third, to check whether discontinuities were unique to the Medicaid threshold, we conducted placebo analyses at income levels above and below 100% FPL. Fourth, because access to care facilitated by Medicaid may affect health, we examined discontinuities in self-reported health status. Additional sensitivity analyses are reported in the Appendix.¹⁷

Limitations

Our study had limitations. First, differences in the characteristics of Medicare beneficiaries above and below the Medicaid threshold limit our ability to isolate effects of the coverage cliff from confounders. We accounted for observable differences by controlling for respondent-level covariates and by running analyses stratified by disability status. Second, small sample sizes may have limited our ability to detect effects for certain outcomes. Third, income in the HRS is self-reported, imputed for respondents who do not report exact values, and thus may be measured with error. However, we were able to identify a sizeable discontinuity in Medicaid enrollment immediately above vs. below Medicaid’s eligibility threshold, suggesting that HRS income data were sufficiently reliable to assess Medicaid eligibility. Fourth, our analyses focus on traditional Medicare and may not generalize to Medicare Advantage, where cost sharing is often lower.¹ Fifth, the gradual closure of the Part D donut hole, which began in 2011, lowered out-of-pocket drug costs for Medicare beneficiaries who do not receive the LIS. We lacked statistical power to compare discontinuities in medication use before and after 2011. However, the LIS still substantially lowers out-of-pocket costs for most Part D enrollees.²⁴

RESULTS

Study sample and population characteristics

Our sample consisted of 18,689 respondent-years (4,602 respondents), which when weighted represented 71,393,444 person-years in the community dwelling population with traditional Medicare and income <200% FPL. Medicare beneficiaries whose income was immediately above vs. below the threshold for Medicaid supplemental coverage were older, had fewer difficulties performing activities of daily living, were more likely to be white, less likely to be disabled, and less likely to have ever been diagnosed with hypertension, cancer, or a behavioral health disorder (Exhibit 1). Because diagnoses for these conditions could have been established before individuals attained Medicaid eligibility, we cannot attribute these differences to Medicaid. To account for these differences, we controlled for the variables in Exhibit 1.

Exhibit 1:

Characteristics of Medicare beneficiaries above vs. below the income eligibility threshold for Medicaid supplemental coverage

Characteristica	Mean below Medicaid thresholdb	Mean above Medicaid thresholdb	Discontinuity above vs. below Medicaid thresholdc
Age, in years	75.1	76.5	1.5**
Male, %	27.5	29.4	2.0
Race/ethnicity, %d
White	65.5	75.5	9.9****
Black	15.1	13.5	−1.6
Hispanic	13.6	8.0	−5.6
Asian	1.4	1.2	−0.2
Other	4.4	1.8	−2.5
Married or living with a partner, %	24.4	23.6	−0.7
Education, in years	10.5	11.3	0.8***
Ever smoked tobacco, %	57.3	57.4	0.1
Disabled, %e	35.3	25.3	−9.9****
End-stage renal disease, %	0.4	0.9	0.5
Used a proxy to respond to the Health and Retirement Study, %	4.3	4.8	0.5
Number of difficulties performing activities of daily livingf	0.8	0.6	−0.2**
Number of difficulties performing instrumental activities of daily livingg	0.6	0.5	−0.1
Likely dementia, %h	14.9	13.0	−1.9
Self-reported health history, % with prior diagnosis of:i
Diabetes	29.4	24.8	−4.6
Hypertension	72.1	65.8	−6.3**
Heart diseasej	32.7	33.4	0.8
Stroke	13.4	12.6	−0.7
Cancerk	19.3	14.6	−4.6*
Chronic lung disease	14.7	17.9	3.2
Behavioral health disorderl	29.5	21.2	−8.2**
Arthritis	72.8	73.9	1.2

Source: Authors’ analyses of the 2008–2016 biennial waves of the Health and Retirement Study linked to Medicare enrollment and claims files from 2007–2016.

Statistical significance of discontinuities is denoted as follows:

^*P<0.10,

^**P<0.05,

^***P<0.01,

^****P<0.001.

^aCharacteristics of Medicare beneficiaries meeting study criteria. When weighted, our sample of 18,689 respondent-years in the HRS represented 71,393,444 person-years in the community dwelling population with traditional Medicare and income <200% FPL.

^bMean of characteristic among respondents with income either above or below the income threshold for Medicaid supplemental coverage (100% FPL), as indicated by the table column, adjusted for income, state fixed effects, and year fixed effects (Appendix).

^cEstimated discontinuity in the relationship between income and each characteristic above vs. below the income threshold for Medicaid supplemental coverage.

^dWe tested the statistical significance of the discontinuity in respondents who were nonwhite (vs. white) above vs. below the Medicaid eligibility threshold.

^eDisability was original reason for Medicare entitlement.

^fNumber of difficulties performing activities of daily living (that is, bathing, dressing, eating, getting into and out of bed, and walking), lagged by one survey wave.

^gNumber of difficulties performing instrumental activities of daily living (that is, using the telephone, taking medications, managing money, shopping for groceries, and preparing meals), lagged by one survey wave.

^hWe used responses to the telephone interview for cognitive status administered in the HRS to assess cognition. We categorized respondents with scores of 0–6, out of 27 possible points, as likely having dementia.

^IIndicators reflect whether a doctor ever diagnosed the respondent with the following health conditions, lagged by one survey wave.

^jThat is, a heart attack, coronary heart disease, angina, congestive heart failure, or other heart problem.

^kExcludes skin cancer.

^lThat is, an emotional, anxiety, or psychiatric disorder.

When we compared Medicare beneficiaries immediately above vs. below Medicaid’s eligibility threshold, we found fewer differences in these characteristics than when we compared Medicare beneficiaries in our sample who did vs. did not receive Medicaid (Appendix Exhibit 9¹⁷). This suggests that our regression discontinuity design mitigated confounding.

Effects on supplemental coverage

We estimated that 73.3% of Medicare beneficiaries whose incomes were just below Medicaid’s eligibility threshold had any supplemental coverage, compared to 47.5% of beneficiaries whose incomes were above this threshold, resulting in a discontinuity in supplemental coverage of 25.8 percentage points (95% CI: −31.3,−20.3;P<0.001; Exhibit 3). Contributing to this supplemental coverage cliff was a discontinuity in Medicaid supplemental coverage of −37.0 percentage points and a discontinuity in private supplemental coverage of 9.1 percentage points.

Exhibit 3:

Adjusted discontinuities and effects of the supplemental coverage cliff on out-of-pocket spending and use of Part A and Part B services^a

		Mean below Medicaid thresholdb		Discontinuity above vs. below Medicaid thresholdc		Effect of supplemental coverage cliffd
Receipt of supplemental coverage:
Medicaid or private supplemental coverage, %		73.3		−25.8****		—
Medicaid supplemental coverage, %		47.3		−37.0****		—
Private supplemental coverage, %		29.4		9.1****		—
Study Outcomes:
Out-of-pocket medical spending, prior 2 years:e
Total spending, $		2025		586**		2288
Total spending >$2,000, %		25.9		8.5***		33.1
Spending on physician office visits, $		225		107**		419
Spending on hospital care, $		134		152***		591
Other medical spending, excl. prescription drugs, $		480		16		62
Outpatient services in physician offices and hospital outpatient departments:
Annual services (total)		28.0		−2.5		−9.5
Annual services, by Berenson-Eggers Type of Service code:f
Evaluation and management		10.3		−1.5***		−5.7
Procedures		4.6		0.1		0.4
Imaging		3.6		−0.4*		−1.5
Tests		6.4		−0.2		−0.8
Other		2.5		−0.3		−1.0
Hospital and skilled nursing facility care:
Annual emergency department visits and observation stays		0.53		−0.05		−0.20
Annual inpatient admissions		0.29		0.05		0.18
Annual inpatient admissions for ambulatory care-sensitive conditions		0.04		−0.005		−0.02
Skilled nursing facility admissions		0.06		0.002		0.01

Source: Authors’ analyses of the 2008–2016 biennial waves of the Health and Retirement Study linked to Medicare enrollment and claims files from 2007–2016.

Statistical significance is denoted as follows:

^*P<0.10,

^**P<0.05,

^***P<0.01,

^****P<0.001.

^aUnless otherwise noted, estimates are based on a sample of 18,689 respondent-years.

^bMean among Medicare beneficiaries whose income was immediately below the threshold for Medicaid supplemental coverage (100% FPL). The Appendix provides details of this calculation.

^cDiscontinuities are differences immediately above vs. below the income threshold for Medicaid supplemental coverage. Shown are discontinuities in supplemental coverage or the study outcome, adjusted for income, respondent characteristics from Exhibit 1, state fixed effects, year fixed effects, and HRS survey weights (see Appendix for details of the regression model used to obtain these estimates).

^dEffects of the supplemental coverage cliff among near-poor Medicare beneficiaries, calculated using instrumental variable methods as described in the methods and Appendix. Effects equal the adjusted discontinuity in the study outcome divided by the absolute value of the adjusted discontinuity in supplemental coverage (25.8 percentage points). Dividing figures in this column by the mean of the outcome just below the Medicaid eligibility threshold gives the effect of the supplemental coverage cliff in relative terms. P-values for effects of the supplemental coverage cliff are the same as for the adjusted discontinuities in study outcomes.

^eSample is limited to HRS survey years. Estimates are based on a sample of 9,599 respondent-years.

^fWe do not report use of durable medical equipment, which accounted for <0.1 of Part B claims.

Effects on study outcomes

We found sizeable discontinuities in out-of-pocket medical spending above vs. below the Medicaid threshold: $586 in higher out-of-pocket spending over the previous 2 years (95% CI: 150,1023;P=0.01; Exhibits 2 and 3), and an 8.5 percentage point higher probability of incurring catastrophic spending over this period (95% CI: 3.7,13.3;P=0.001). Using instrumental variables to estimate effects of the supplemental coverage cliff, we found that near-poor beneficiaries exposed to this coverage cliff incurred $2,288 in additional out-of-pocket spending (over 2 years) and were 33.1 percentage points more likely to incur catastrophic spending.

Exhibit 2:

Out-of-pocket medical spending above and below the Medicaid eligibility threshold

Source: Authors’ analyses of the 2008–2016 biennial waves of the Health and Retirement Study linked to Medicare enrollment and claims files from 2007–2016.

Notes: Scatterplot shows unadjusted levels of out-of-pocket medical spending by the income of Medicare beneficiaries. Estimates adjusted for HRS survey weights. The fitted regression lines and 95% confidence intervals (superimposed on the scatterplot) represent the relationship between income and out-of-pocket spending above vs. below the eligibility threshold for Medicaid supplemental coverage (100% FPL). The vertical distance between the fitted regression lines at 100% FPL gives the unadjusted discontinuity in out-of-pocket medical spending. We excluded respondents whose income was within ±4 percentage points of this threshold. See the Appendix for plots of all study outcomes and of the receipt of supplemental coverage. Estimates are based on 9,599 respondent-years.

Medicare beneficiaries’ use of outpatient E&M services was lower immediately above vs. below the Medicaid threshold (discontinuity: −1.5 annual services; 95% CI: −2.5,−0.5;P=0.003; Exhibit 3). Consequently, we estimated that beneficiaries affected by the coverage cliff used 5.7 fewer outpatient E&M services per year (a 55% reduction from mean use below the Medicaid threshold). We did not detect discontinuities in emergency department visits and observation stays, inpatient admissions, or skilled nursing facility admissions.

Exhibit 4:

Adjusted discontinuities in prescription drug subsidies and effects on spending and utilization

		Mean below Medicaid thresholda		Discontinuity above vs. below Medicaid thresholdb		Effect of the Part D Low-Income Subsidy (LIS) enrollment cliffc
Enrolled in Medicare Part D, %d		77.0		−4.1		—
Among Part D enrollees:e
Receipt of the Part D Low-Income Subsidy (LIS), %		75.7		−11.5****		—
Annual out-of-pocket spending on Part D-covered drugs, $		197		94***		813
Annual LIS payments for Part D-covered drugs, $		1265		−326***		−2833
Annual Part D claims:
Total		46.8		−7.0***		−60.5
Chronic disease medicationsf		26.0		−2.8**		−24.4

Source: Authors’ analyses of the 2008–2016 biennial waves of the Health and Retirement Study linked to Medicare enrollment and claims files from 2007–2016.

Statistical significance is denoted as follows:

^*P<0.10,

^**P<0.05,

^***P<0.01,

^****P<0.001.

^aMean among Medicare beneficiaries whose income was immediately below the threshold for Medicaid supplemental coverage (100% FPL).

^bDiscontinuities are differences immediately above vs. below the income threshold for Medicaid supplemental coverage. Shown are discontinuities in in Part D coverage, prescription drug subsidies, spending, and utilization, adjusted for the relationship between income, respondent characteristics from Exhibit 1, state fixed effects, and year fixed effects, and HRS survey weights.

^cEffects of the Part D Low-Income Subsidy (LIS) enrollment cliff at 100% FPL. Effects were calculated using instrumental variable methods and equal the adjusted discontinuity in the study outcome divided by the absolute value of the adjusted discontinuity in LIS enrollment (11.5 percentage points). Dividing figures in this column by the mean of the outcome just below the Medicaid eligibility threshold gives the effect of the LIS enrollment cliff in relative terms. P-values for effects of the LIS enrollment cliff are the same as for the adjusted discontinuities in study outcomes.

^dEstimates based on 18,689 respondent-years.

^eUnless otherwise noted, estimates limited to respondents continuously enrolled in Medicare Part D in the study year (or until death for decedents) and were based on 12,385 respondent-years.

^fCount of part D claims for drugs primarily used to treat one of the following chronic conditions: asthma, chronic obstructive pulmonary disease, depression, diabetes, heart failure, hyperlipidemia, hypertension, ischemic heart disease or acute myocardial infarction, or glaucoma. Analyses limited to Part D enrollees with 2 or more of these conditions (N=10,934 respondent-years). See Appendix for a description of the drug classes that we analyzed.

We did not detect a discontinuity in Part D coverage above vs. below the Medicaid threshold. However, among Part D enrollees, exceeding this threshold was associated with lower LIS enrollment (discontinuity of −11.5 percentage points; 95% CI: −16.9,−6.1;P<0.001; Exhibit 4). We estimated that beneficiaries affected by this LIS enrollment cliff incurred $813 in additional out-of-pocket spending on prescription drugs (413% higher than below the Medicaid threshold), filled 60.5 fewer prescriptions per year (129% fewer than below the threshold), and filled 24.4 fewer chronic disease medications per year (94% fewer).

Estimates stratified by disability status

We found similar effects of the coverage cliff on out-of-pocket spending and outpatient and prescription drug use when we stratified our analyses by Medicare beneficiaries’ disability status. However, we found a larger discontinuity in supplemental coverage among disabled vs. non-disabled beneficiaries (47.1 vs. 14.9 percentage points), and we only detected a cliff in LIS enrollment among non-disabled Part D enrollees (Appendix Exhibits 10–11¹⁷).

Supplementary analyses

We found no evidence to suggest that respondents manipulated income to attain Medicaid eligibility (Appendix Exhibits 12–14¹⁷). Compared to near-poor Medicare beneficiaries, those with incomes <100% FPL were more likely to receive other means-tested assistance such as food stamps (Appendix Exhibit 15¹⁸). However, these programs do not affect out-of-pocket health care costs. For outcomes where we found discontinuities at the Medicaid threshold, we did not detect discontinuities at other income levels, giving us confidence that the effects we detected were attributable to the coverage cliff (Appendix Exhibit 16¹⁷). We did not detect a discontinuity in self-reported health status (Appendix Exhibit 17¹⁷). Appendix Exhibits 18–19¹⁷ report the results of other sensitivity analyses.

DISCUSSION

We studied the consequences of cost sharing among near-poor Medicare beneficiaries by examining a cliff in supplemental coverage at the income eligibility threshold for Medicaid. Near-poor beneficiaries, whose incomes exceed this threshold, are nearly 26 percentage points less likely to have any supplemental coverage and consequently must pay Medicare’s full cost sharing out-of-pocket. We found that beneficiaries affected by this supplemental coverage cliff incurred higher out-of-pocket costs and used fewer outpatient services, but we did not detect effects on hospital or skilled nursing facility use. We also found a spillover of this coverage cliff on prescription drug use, due in part to lower LIS enrollment among near-poor beneficiaries who qualify for this benefit.

Our results demonstrate that health care use among near-poor Medicare beneficiaries is highly sensitive to out-of-pocket costs. A common economic measure of sensitivity to out-of-pocket costs is the price elasticity of demand, which is calculated as a percent change in use divided by a percent change in the out-of-pocket price of care. In prior studies of cost sharing, including the RAND Health Insurance Experiment, researchers estimated price elasticities of demand of approximately −0.2, meaning that a 10% increase in prices reduces use by 2%.^10,25 Our analyses of outpatient evaluation and management visits suggest that sensitivity to prices among near-poor Medicare beneficiaries is twice as large, with a price elasticity of demand of approximately −0.4.²⁶ This greater sensitivity to out-of-pocket prices underscores the economic burden of Medicare’s cost sharing for near-poor beneficiaries who lack supplemental coverage and demonstrates the importance of Medicaid in mitigating barriers to care related to cost.

Our findings were similar when we separately analyzed disabled and non-disabled Medicare beneficiaries, with two exceptions. First, the supplemental coverage cliff was larger among disabled vs. non-disabled individuals, highlighting the importance of Medicaid as a supplemental insurer for disabled beneficiaries and institutionally linked pathways to Medicare and Medicaid enrollment for low-income and disabled populations.²³ Second, among disabled Part D enrollees, we found lower medication use, without a discontinuity in LIS enrollment, above Medicaid’s eligibility threshold. This finding suggests that mechanisms other than receipt of the LIS, such as the improved ability to afford prescriptions when Medicaid covers Part A and B cost sharing, may also affect medication use.

Policy implications

Our findings have several policy implications. First, expanding Medicaid supplemental coverage to near-poor Medicare beneficiaries would lessen the financial burden of Medicare’s cost sharing and increase beneficiaries’ use of some outpatient services. The Medicare Payment Advisory Commission has evaluated proposals to expand eligibility for Medicaid supplemental coverage to 150% FPL,²⁷ and some analysts have proposed expanding eligibility to 200% FPL.¹² Expanding Medicaid supplemental coverage to 150% FPL would align Medicaid eligibility with the LIS.²⁷ However, a broader expansion to 200% FPL would help to fill a gap in supplemental coverage that persists up to this income level and lessen substantial out-of-pocket cost burdens that are experienced by Medicare beneficiaries in the 100–200% FPL income range.^12,13

Second, our results underscore that policy makers should design Medicaid benefits to mitigate cliff effects present in the current Medicaid program. For example, under either expansion scenario, cost sharing subsidies could be provided to the near poor on a sliding scale (e.g., gradually tapering the proportion of cost sharing that Medicaid pays for those with higher incomes) or structured so that individuals pay no more than a fixed proportion of their income on Medicare costs (thus, people with higher incomes would have a higher out-of-pocket limit).

Policy makers considering an expansion of Medicaid supplemental coverage would need to address other programmatic issues. Specifically, the expansion proposals summarized here would cover Medicare beneficiaries with incomes >100% to ≤135% FPL, who currently qualify for limited Medicaid benefits that pay the Part B premium. These premium subsidies could be incorporated into an expanded Medicaid benefit and provided on a sliding scale up to the new Medicaid eligibility threshold.^12,13,27 Consequently, Medicaid’s existing Part B premium subsidy programs, which have low participation rates,¹⁶ would be eliminated. Policy makers would also need to decide what share of costs would be federally financed and whether Medicaid benefits for Medicare beneficiaries would continue to be administered by states or instead administered by the federal government (as is the case for the LIS).²⁷

Third, our findings highlight an opportunity to increase LIS enrollment among near-poor beneficiaries who qualify for this benefit, because beneficiaries who do not receive the LIS are burdened with higher out-of-pocket costs that contribute to lower medication use. Medicare beneficiaries automatically receive the LIS when they enroll in Medicaid, and auto-enrollment is the main pathway into the LIS.^15,16 Therefore, to increase LIS enrollment, policy reforms could focus on increasing take-up of Medicaid benefits currently offered to near-poor beneficiaries (i.e., subsidies for the Part B premium) and maximizing take-up of expanded Medicaid supplemental coverage (if implemented).

Comparisons to prior research

Results of our study are consistent with prior research on supplemental insurance for Medicare beneficiaries, which has linked receiving private supplemental coverage to increased use of physician services²⁸ and Medicaid to a lower probability of avoiding physician visits due to cost.²⁹ Our results are also consistent with evidence that near-poor Medicare beneficiaries are more likely to experience cost-related medication non-adherence.³⁰ However, our findings differ from several other studies that linked higher cost sharing for outpatient care and prescription drug use to increased hospital use.^9,25,31 We did not find an effect of the supplemental coverage cliff on hospital use, although this may be because Medicare beneficiaries without supplemental coverage also face higher cost sharing for hospital care, whereas prior studies examined variation in cost sharing that was limited to outpatient care and prescription drugs.^9,25,31

Conclusions

Near-poor Medicare beneficiaries face substantially higher out-of-pocket costs because they are ineligible for Medicaid supplemental coverage but frequently lack private supplemental insurance. We found that this supplemental coverage cliff was associated with lower use of some outpatient services and with spillovers on prescription drug use. Expanding Medicaid supplemental coverage to this near-poor population, coupled with policies to increase enrollment in the LIS, would lessen cost-related barriers to health care use among near-poor Medicare beneficiaries.