Loss of Subsidized Drug Coverage and Mortality Among Low-Income Medicare Beneficiaries

Abstract

Background:

Thirteen million Medicare beneficiaries receive the Low-Income Subsidy (LIS), which reduces cost-sharing in Medicare Part D. Losing LIS may impede medication access and affect mortality.

Methods:

Using 2015–2023 Medicare data, we identified dual-eligible Medicare-Medicaid beneficiaries, who automatically receive LIS, and calculated annual Medicaid and LIS loss rates. To examine the relationship between LIS loss and mortality, we leveraged a natural experiment arising from the relationship between the timing of Medicaid disenrollment and subsequent LIS loss. We compared beneficiaries disenrolling from Medicaid in January-June, who kept LIS through December (6–11 additional months), to those disenrolling in July-December, who kept LIS through the following December (12–17 additional months). Among individuals disenrolling from Medicaid during 2015–2017, we examined cumulative mortality 7–17 months post-disenrollment, when earlier disenrollees were more likely to lose LIS.

Results:

The sample included 969,606 individuals with early (January-June) and 920,158 with late (July-December) Medicaid disenrollment. Those disenrolling from Medicaid earlier averaged 13.6 cumulative months of LIS in the 17 months after disenrollment, versus 15.3 months for later disenrollees. Seventeen months after Medicaid disenrollment, cumulative mortality was higher among early disenrollees (78.3 per 1000) versus later disenrollees (75.3 per 1000), a difference of 3.0 deaths per 1000 (95% CI: 2.1,3.9). Mortality increases were amplified among those in the highest quintile of baseline Part D spending (5.6 deaths per 1000; 95% CI: 3.3,7.9) and users of cardiovascular disease, chronic lung disease, and HIV medications.

Conclusions:

Loss of drug subsidies following Medicaid disenrollment is associated with increased mortality among low-income Medicare beneficiaries.

INTRODUCTION

Consistent access to medications can improve health and save lives.^1–4 Since 2006, individuals enrolled in Medicare have had access to prescription drug coverage through Medicare Part D, whose introduction has been linked to increased medication use and lower mortality from chronic conditions.^5,6 However, barriers to medication access remain due to substantial cost-sharing in the Part D program, which includes an annual deductible, copayments, and coinsurance.^7,8 In 2022, 1 in 7 Medicare beneficiaries did not fill a prescription due to cost concerns,⁹ and a disproportionate share of those with low incomes, multimorbidity, or a disability reported difficulty affording medications.^10,11

The Low-Income Subsidy (LIS), established alongside Part D, reduces cost barriers for beneficiaries with limited incomes and assets. Most LIS enrollees have zero-dollar premiums to enroll in Part D and make minimal copayments for Part D drugs (≤$4.90 for generics and ≤$12.15 for branded drugs in 2025).^12,13 Research has linked the LIS to increased medication adherence for chronic conditions.^14–16 The Social Security Administration (SSA), which administers LIS, estimates that this subsidy saves enrollees an average of $6,200 annually.¹⁷

However, administrative hurdles put Medicare beneficiaries at risk of losing the LIS. Approximately 85% of LIS enrollees receive this subsidy automatically because they are dually eligible for and enrolled in Medicare and Medicaid.^18,19 For these dual-eligible individuals, LIS loss may be precipitated by Medicaid disenrollment, which can occur when states conduct Medicaid eligibility redeterminations.²⁰ Beneficiaries who lose LIS face higher out-of-pocket medication costs and may lose Part D coverage if they cannot pay the monthly premium.¹⁹ A sudden disruption in access to affordable medications may adversely affect health, especially because the dual-eligible population is socially vulnerable and medically complex.²¹ The frequency with which LIS loss occurs among dual-eligible beneficiaries and its association with health outcomes remain unclear.

Therefore, this study had two objectives. First, we estimated the share of dual-eligible beneficiaries who disenrolled from Medicaid and LIS annually. Second, we examined the association between LIS loss with mortality, leveraging a natural experiment wherein the month of Medicaid disenrollment affected when beneficiaries lost LIS.

METHODS

Study design

We studied a natural experiment arising from the relationship between Medicaid enrollment and LIS eligibility. (For program details, see Online Appendix A.) Medicare beneficiaries with Medicaid are automatically eligible for LIS¹⁸ and continue to receive this subsidy even after disenrolling from Medicaid. However, the exact timing of Medicaid disenrollment affects how long individuals retain LIS before losing automatic eligibility. Those disenrolling from Medicaid between January and June retain LIS through December (6–11 additional months), while those disenrolling between July and December retain LIS through the following December (12–17 additional months).^19,22 Thereafter, beneficiaries lose LIS unless they re-enroll in Medicaid or apply for LIS directly via the SSA.

States conduct Medicaid redeterminations throughout the year, leading to disenrollments in all months. Because the staggered timing of Medicaid disenrollment affects when individuals lose LIS but is unlikely to be correlated with confounders, we leverage it as a natural experiment to examine the relationship between LIS loss and mortality. We compared Medicare beneficiaries who disenrolled from Medicaid earlier (January-June) vs. later (July-December) in the same year. We followed individuals for 24 months, focusing on mortality in the 11-month window (i.e., months 7–17 post-Medicaid disenrollment) when LIS losses were greater for those with early vs. late Medicaid disenrollment (Online Appendix B).

Data

Our primary dataset was the Medicare Beneficiary Summary File (MBSF) for the years 2014–2023 (including a one-year lookback period for baseline variables). The MBSF reports beneficiary demographics, Medicare enrollment, monthly Medicaid and LIS enrollment, and death dates. We also used Part D event files to measure drug spending and identify individuals filling prescriptions in specific therapeutic classes at baseline (i.e., during the 12 months preceding Medicaid disenrollment). Among beneficiaries with traditional Medicare (TM) at baseline, we used diagnoses in TM claims to construct CMS Hierarchical Condition Category (HCC) risk scores.

Study Samples

To examine Medicaid and LIS loss during 2015–2023, we analyzed annual cohorts of beneficiaries enrolled in Medicare (Parts A, B, and D) and Medicaid in January. We limited the sample to beneficiaries who kept Part A while alive to observe Medicaid disenrollment.

To examine outcomes associated with LIS loss, we analyzed a 100% sample of Medicare beneficiaries who disenrolled from Medicaid for ≥1 month during 2015–2017, ensuring a 24-month follow-up period unaffected by the COVID-19 Public Health Emergency (PHE). To measure baseline characteristics and medication fills, we required 12 months of continuous enrollment in Medicare (Parts A, B, and D) at baseline. To observe deaths in MBSF, we required continuous Part A enrollment (while alive) during the 24 months after Medicaid disenrollment. We excluded individuals who moved states, which can affect Medicaid eligibility, and randomly selected one disenrollment episode for those losing Medicaid multiple times. Online Appendix C details our inclusion criteria and Online Appendix D defines key study variables. This study was approved by the Harvard Longwood Campus Institutional Review Board, which waived the requirement for informed consent because this was a retrospective study of de-identified data.

Outcome & Covariates

Our primary outcome was all-cause mortality, and the secondary outcome was the number of Part D medications filled per month. We included the following beneficiary-level covariates: age at Medicaid disenrollment (categorized in 10-year increments), sex, race and ethnicity, original reason for Medicare entitlement (e.g., age, disability, end-stage renal disease), enrollment in Medicare Advantage (MA) vs. TM in the month before disenrollment, and receipt of full vs. partial Medicaid before disenrollment. Full Medicaid covers individuals with very low incomes and assets and pays for long-term care in addition to most Medicare out-of-pocket costs; partial Medicaid covers individuals with modestly higher incomes and assets but only assists with certain Medicare out-of-pocket costs.²¹

Statistical Analyses

Annual Medicaid and LIS loss, 2015–2023

We examined annual rates of Medicaid and LIS loss (for ≥1 month) from 2015–2023. This included the period March 2020-March 2023, when Medicaid redeterminations were paused under the COVID-19 PHE,²³ and April-December 2023, when redeterminations resumed.

Association between LIS loss and mortality

Our primary analyses compared individuals who disenrolled from Medicaid between January-June (early) vs. July-December (late) during 2015–2017. We first compared monthly LIS enrollment for beneficiaries who disenrolled from Medicaid earlier vs. later in the same year. We also examined monthly enrollment in Part D and re-enrollment in Medicaid.

Second, we compared cumulative survival for the early vs. late disenrollment groups over 3 post-disenrollment periods: 1) months 1–6, when both groups kept LIS; 2) months 7–17, when LIS losses were greater in the early vs. late disenrollment groups; and 3) months 18–24, when LIS receipt declined in both groups. Because both groups retained LIS in the first 6 months after Medicaid disenrollment, we analyzed this 6-month window as a placebo period when mortality trends would be expected to evolve similarly between the groups.

Third, we estimated regression-adjusted differences in cumulative mortality, using an intention-to-treat approach that compared beneficiaries with early vs. late Medicaid disenrollment. We fitted person-month-level linear regressions that modeled mortality as a function of early vs. late disenrollment, fixed effects for post-disenrollment months, and their interaction. We focused on cumulative mortality through month 17 post-Medicaid disenrollment, as this marked the end of the period when LIS losses differed for early vs. late disenrollees. However, mortality differences could fluctuate during the study period due to lagged effects of LIS loss and subsequent LIS losses for later Medicaid disenrollees.

Models were adjusted for covariates described above, fixed effects for birth month (to account for differences in Part D cost-sharing for individuals aging into Medicare in different months),² state and year of disenrollment, and two-way interactions between these variables with indicators for months since disenrollment. To adjust for differences in seasonal exposure timing (e.g., later disenrollees reaching winter sooner), we included a continuous person-month-level measure of cumulative winter exposure through each post-disenrollment month. We used robust standard errors accounting for intra-person clustering. Confidence interval widths were not adjusted for multiplicity and may not be used in place of hypothesis testing. Additional details are in Online Appendix E.

In subgroup analyses, we examined beneficiaries by baseline Part D spending quintile and those who filled ≥30 days of medications in eight therapeutic classes for chronic disease management (e.g., cardiovascular disease, HIV; Online Appendix D). We also stratified the sample according to baseline enrollment in full vs. partial Medicaid, MA vs. TM, and for those with TM, by quintile of the HCC risk score. Finally, we examined changes in Part D prescription fills (while beneficiaries were alive) during months when LIS loss differed between early and late Medicaid disenrollees.

Our study design assumed that Medicaid disenrollment timing is uncorrelated with confounders. We tested this assumption by examining whether baseline characteristics were balanced between early vs. late disenrollees, verifying that mortality differences were minimal during the 6 months between Medicaid disenrollment and LIS loss, and assessing bias from state-level factors and seasonality. To examine whether findings were sensitive to unmeasured state factors, we conducted a “leave-one-out” analysis that iteratively excluded one state at a time. To mitigate timing differences in seasonal exposures, we conducted a sensitivity analysis of individuals disenrolling from Medicaid within a narrower timeframe (April-September).

RESULTS

Annual Medicaid and LIS loss, 2015–2023

In each year from 2015 to 2019, 6.7% to 7.4% of beneficiaries lost Medicaid and 1.9% to 2.5% lost LIS for ≥1 month. Medicaid disenrollment declined to 1.2% in 2021, while LIS loss declined with a one-year lag (e.g., 0.9% in 2022). Medicaid disenrollment increased to 7.4% in 2023 when the COVID-19 PHE ended (Appendix Figure 8).

Association between LIS loss and mortality

Sample characteristics:

We analyzed 1,889,764 beneficiaries who disenrolled from Medicaid for ≥1 month during 2015–2017: 969,606 disenrolling between January and June (early), and 920,158 disenrolling between July and December (late) (Table 1). Baseline characteristics were balanced across early and late disenrollment groups, with standardized mean differences <0.1.

Table 1:

Characteristics of beneficiaries with Medicaid disenrollment, 2015–2017

Characteristic at baselinea	Early Medicaid disenrollment (January-June) (n=969,606)	Late Medicaid disenrollment (July-December) (n=920,158)	Standardized mean differenceb
Age, %c
<65 years	45.9%	46.5%	0.011
65–74 years	30.5%	31.4%	0.019
75–84 years	16.0%	15.2%	0.022
≥85 years	7.5%	6.9%	0.023
Sex, %
Female	55.7%	55.8%	0.002
Male	44.3%	44.2%	0.002
Race and ethnicity, %d
Non-Hispanic White	55.4%	54.3%	0.021
Non-Hispanic Black	21.0%	21.2%	0.006
Hispanic	17.9%	18.2%	0.007
Other or unknown	5.8%	6.3%	0.023
Original reason for Medicare entitlement, %
Age	39.4%	39.9%	0.010
Disability	58.9%	58.5%	0.008
End-stage renal disease	0.6%	0.6%	0.003
Disability and end-stage renal disease	1.1%	0.9%	0.012
Medicaid coverage type, %
Full Medicaid	50.8%	50.9%	0.002
Partial Medicaid	49.2%	49.1%	0.002
Medicare coverage type, %
Medicare Advantage	37.8%	38.9%	0.024
Traditional Medicare	62.2%	61.1%	0.024
HCC risk score, mean (SD)e	1.50 (1.43)	1.47 (1.41)	0.024
Baseline medication use, % filling prescriptions by therapeutic class,f
Diabetes drugs	26.4%	26.1%	0.006
Antithrombotic drugs	15.0%	14.5%	0.014
Cardiovascular disease drugs	67.3%	66.5%	0.017
Lipid-lowering drugs	44.2%	44.0%	0.004
Immunosuppressants	3.3%	3.3%	0.001
Drugs for obstructive airway disease	29.8%	29.7%	0.004
Neuropsychiatric drugs	57.8%	57.2%	0.011
Antiretroviral drugs	1.3%	1.3%	<0.001

Characteristics of Medicare beneficiaries with early vs. late Medicaid disenrollment in 2015–2017. Disenrollment is defined as loss of Medicaid for ≥1 month. Estimates pool beneficiaries with full and partial Medicaid at baseline.

Appendix Figure 9 shows the distribution of beneficiaries with early vs. late Medicaid disenrollment by state.

Appendix Figure 10 summarizes overall covariate balance among beneficiaries disenrolling from Medicaid early vs. later in the same year. Appendix Table 3 shows characteristics separately for beneficiaries with full vs. partial Medicaid at baseline.

^aUnless otherwise noted, the baseline period is the month immediately preceding Medicaid disenrollment.

^bDifference in means or proportions of a characteristic between groups with early vs. late Medicaid disenrollment, divided by the standard deviation of the characteristic (pooled across the early and late disenrollment groups).

^cRegression analyses adjusted for age categorized as ≤25, 25–34, 35–44, 45–54, 55–64, 65–74, 75–84, or ≥85 years.

^dAssessed from the RTI race and ethnicity variable in the Medicare Beneficiary Summary File. The RTI variable is based on Social Security Administration enrollment data and an imputation algorithm that identifies additional Hispanic and Asian beneficiaries.

^eHCC is the Hierarchical Condition Category risk score, constructed using Medicare enrollment and claims data in the 12 months preceding Medicaid disenrollment. Higher scores indicate higher predicted spending in the subsequent year. Because Medicare Advantage encounter data were unavailable or incomplete in some years, HCC risk scores were calculated for beneficiaries with continuous enrollment in Traditional Medicare in the 12 months preceding Medicaid disenrollment.

^fAmong prescription medications covered by Medicare Part D and filled in the 12 months preceding Medicaid disenrollment. See Appendix D for a list of drugs by therapeutic class.

Coverage changes following Medicaid disenrollment:

LIS losses began 7 months after Medicaid disenrollment for early disenrollees and after 13 months for later disenrollees (Figure 1A). Declines in Part D enrollment lagged LIS loss by approximately 3 months (Figure 1B), aligning with the standard grace period for premium nonpayment before Part D coverage is terminated.²² Seventeen months after Medicaid disenrollment, early vs. late disenrollees averaged 13.6 vs. 15.3 cumulative months of LIS coverage, and 15.6 vs. 16.1 cumulative months of Part D enrollment, respectively. Medicaid re-enrollment rates were similar between the early and late disenrollment groups (Figure 1C). For example, 12 months after disenrolling from Medicaid, 50.8% of those who disenrolled early and 47.6% of those who disenrolled later had regained Medicaid.

Figure 1:

Changes in enrollment in the Low-Income Subsidy, Medicare Part D, and Medicaid among beneficiaries with Medicaid disenrollment, 2015–2017

Panel A: Enrollment in the Low-Income Subsidy

Panel B: Enrollment in Medicare Part D

Panel C: Enrollment in Medicaid

Figures depict enrollment in the Low-Income Subsidy (Panel A), Part D (Panel B), and Medicaid (Panel C), by month following Medicaid disenrollment among beneficiaries with early vs. late Medicaid disenrollment during 2015–2017. Proportions are out of months while beneficiaries were alive and enrolled in Medicare Part A. Estimates pool beneficiaries with full and partial Medicaid at baseline. Appendix Figure 11 shows enrollment separately for beneficiaries with full vs. partial Medicaid at baseline

Mortality:

During the first 6 months after Medicaid disenrollment, unadjusted survival rates were similar in the early and late Medicaid disenrollment groups (Figure 1). Survival rates were lower for early disenrollees beginning in month 7 post-disenrollment, coinciding with LIS losses in this group, and remained lower through month 17 post-disenrollment.

In regression-adjusted analyses, cumulative mortality did not differ between early vs. late Medicaid disenrollees at 3 and 6 months post-disenrollment (e.g., mortality difference at 6 months: 0.5 deaths per 1000; 95% CI: −0.2,1.2) (Figure 3). Subsequently, mortality increased in the early vs. late disenrollment groups. Seventeen months after disenrollment, cumulative mortality was higher among early vs. late disenrollees (78.3 vs. 75.3 per 1000, difference of 3.0 deaths per 1000, 95% CI 2.1, 3.9).

Figure 3:

Adjusted differences in cumulative mortality, beneficiaries with early vs. late Medicaid disenrollment, 2015–2017

Cumulative mortality per 1000 beneficiaries

Figure shows regression-adjusted differences in cumulative mortality between beneficiaries with early (January-June) vs. late (July-December) Medicaid disenrollment during 2015–2017. Cumulative mortality estimated through a given month following Medicaid disenrollment. Estimates obtained from a person-month-level linear regression that modeled cumulative mortality as a function of early vs. late Medicaid disenrollment, fixed effects for post-disenrollment months, and their interaction. Models adjusted for baseline age, sex, race and ethnicity, original reason for Medicare entitlement, enrollment in full vs. partial Medicaid, Medicare coverage type (Medicare Advantage vs. Traditional Medicare), month of birth, fixed effects for state and year of Medicaid disenrollment, and two-way interactions between these covariates and indicators for months following Medicaid disenrollment. To account for seasonality, estimates were also adjusted for cumulative winter exposure at the beneficiary-month level (total months in December-February accrued through each post-disenrollment month). Whiskers denote 95% confidence intervals. Confidence interval widths were not adjusted for multiplicity and may not be used in place of hypothesis testing. See Appendix Table 4 for regression estimates.

Mortality increases were pronounced among beneficiaries in the highest quintile of baseline Part D spending and smaller in the lower quintiles (Figure 4). Seventeen months after Medicaid disenrollment, mortality in the early disenrollment group was 5.6 deaths per 1000 higher within the highest spending quintile (95% CI: 3.3,7.9) and 1.7 deaths per 1000 higher within the lowest quintile (95% CI: 0.2,3.1). Mortality increases were also amplified among those who, at baseline, filled prescriptions for managing cardiovascular disease, chronic lung disease, and HIV; among those initially enrolled in full Medicaid; and among those in the highest HCC risk score quintile (among TM enrollees) (Appendix Figures 12–14). Enrollees in TM vs. MA at baseline experienced similar mortality increases (Appendix Figure 15).

Figure 4:

Adjusted differences in cumulative mortality by baseline Part D spending quintile

Cumulative mortality per 1000 beneficiaries

Figure depicts regression-adjusted differences in cumulative mortality for beneficiaries with early (January-June) vs. late (July-December) Medicaid disenrollment, stratified by quintile of baseline Part D spending. Baseline Part D spending was measured in the 12 months before Medicaid disenrollment. In the 12 months before disenrollment, per person spending was $0 to $154 in quintile 1; $155 to $774 in quintile 2; $775 to $2,426 in quintile 3; $2,427 to $6,334 in quintile 4; and $6,335 to >$20,000 in quintile 5 (highest value suppressed to preserve confidential data). Separate models were fitted to each spending quintile and each model used the same specification as described in the notes to Figure 3. Whiskers denote 95% confidence intervals. Confidence interval widths were not adjusted for multiplicity and may not be used in place of hypothesis testing. Due to space limits, estimates for month 18 after Medicaid disenrollment are not plotted. See Appendix Table 4 for regression estimates.

The number of medications filled through Part D declined in months when beneficiaries disenrolling from Medicaid early were less likely to have LIS. Cumulatively from months 7 to 17 post-disenrollment, early disenrollees filled an average of 1.2 fewer prescriptions than later disenrollees while they were alive. Those in the lowest Part D spending quintile filled 0.6 fewer prescriptions while alive, and those in the highest quintile averaged 2.6 fewer (Appendix Figure 16).

Mortality increases were detected across all leave-one-out analyses, although the increases were smaller when dropping certain states (e.g., Indiana or California) and larger when dropping others (e.g., Massachusetts or New York) (Appendix Figure 17). The sensitivity analysis of beneficiaries who disenrolled from Medicaid between April and September produced similar results as our main analysis (Appendix Figure 19).

DISCUSSION

In this national study of low-income Medicare beneficiaries, loss of LIS after Medicaid disenrollment was associated with increased all-cause mortality. Specifically, medication filling declined, and mortality increased, during the 11-month window (7–17 months after Medicaid disenrollment) when individuals who disenrolled from Medicaid earlier vs. later had fewer guaranteed months of LIS coverage. Mortality increases were greatest among beneficiaries with high baseline Part D spending, a group facing substantially higher out-of-pocket costs without LIS. Mortality increases were also pronounced among individuals in subgroups with high clinical risks, including those using medications to manage cardiovascular disease, chronic lung disease, and HIV.

This study contributes to research examining the unintended consequences of cost-sharing for prescription drugs and preventive care on health outcomes.^10,24 Prior studies linked higher cost sharing for prescription drugs to decreased medication adherence,^25,26 increased hospitalizations,^25,27,28 and higher mortality;^2,25 these negative consequences were amplified in medically and socioeconomically vulnerable populations.^26–28 Furthermore, research demonstrates that eliminating medication cost-sharing reduces adverse outcomes from healthcare-amenable causes.¹ This study provides new evidence linking the loss of prescription drug subsidies that reduce Part D cost-sharing to reductions in medication use and increased mortality among low-income Medicare beneficiaries.

Our estimates imply that in 2015–2017 there were approximately 2,909 more deaths (=969,606 × 3.0 deaths/1000) among individuals who disenrolled from Medicaid earlier in the calendar year (leading to earlier automatic LIS loss). Because this calculation only accounts for excess mortality associated with earlier Medicaid loss, the excess mortality from all LIS losses during this period is likely larger. Amid increasing Medicaid disenrollments due to the unwinding of the PHE’s continuous Medicaid eligibility policy, our findings highlight concerns that ensuing LIS losses could further reduce access to affordable medications and increase the risk of adverse health outcomes.

Although we could not determine why beneficiaries disenrolled from Medicaid, we found that one-half of those who disenrolled from Medicaid re-enrolled within 12 months. These findings are consistent with other studies^23,29,30 and suggest that administrative factors, rather than eligibility loss, frequently contribute to coverage interruptions. Therefore, this study highlights the importance of reducing administrative burdens that may disrupt Medicaid—and consequently LIS—enrollment. Most LIS enrollees receive this subsidy automatically as a condition of having Medicaid, making continued LIS enrollment dependent on maintaining Medicaid coverage. However, the Medicaid renewal process is complex for Medicare beneficiaries, requiring them to provide documentation of income and assets at least annually. Policies to support continuous Medicaid coverage could help individuals maintain drug subsidies connected to Medicaid. Such policies could include simplifying Medicaid redeterminations (e.g., eliminating asset tests), providing continuous coverage for longer periods (e.g., one year or longer), or verifying eligibility with data from other means-tested programs.³¹

This study had limitations. First, beneficiaries who disenroll from Medicaid earlier vs. later may differ in unmeasured ways that are associated with mortality. Although we found minimal differences between these groups on observable characteristics, there may be residual confounding, limiting our ability to definitively establish causality. Second, timing differences in seasonal exposures could influence mortality. We adjusted for cumulative seasonal exposure and conducted a sensitivity analysis using a narrower timeframe (to reduce seasonal timing differences), which produced similar results. Third, some individuals lost Part D, and we could not observe whether they paid cash or used alternative drug coverage to fill prescriptions. However, in separate analyses of the Medicare Current Beneficiary Survey, we estimated that only 12% of Medicare beneficiaries who lost LIS had alternative drug coverage. Fourth, we did not examine outcomes such as hospitalizations that may reflect adverse health events, underscoring a need for further research to understand other consequences of LIS loss.

CONCLUSION

Loss of prescription drug subsidies following Medicaid disenrollment is associated with increased mortality among low-income Medicare beneficiaries, particularly among individuals with higher baseline drug spending and those using medications to manage cardiovascular disease, chronic lung disease, and HIV. Efforts to increase continuous Medicaid and LIS coverage may help protect the health of low-income populations through access to affordable medications.

Figure 2:

Survival among Medicare beneficiaries with Medicaid disenrollment, 2015–2017

Figure depicts unadjusted proportions of Medicare beneficiaries who remained alive by month from Medicaid disenrollment. Proportions are out of 969,606 beneficiaries with early Medicaid disenrollment (January-June) and 920,158 beneficiaries with late Medicaid disenrollment (July-December). Includes beneficiaries with full and partial Medicaid at baseline