Hospitalizations and Mortality Among Medicare Beneficiaries at Cardiovascular Risk: Disruptions and Recovery from the COVID-19 Pandemic

Abstract

Background:

The COVID-19 pandemic led to disruptions in cardiovascular care in mid-2020, but less is known about how patterns of care and clinical outcomes changed for older adults with cardiovascular (CV) risk factors and/or established cardiovascular disease (CVD) in the later stages of the pandemic.

Objectives:

To identify changes in utilization of healthcare services and all-cause mortality among Medicare beneficiaries with CV risk factors or CVD in the “late pandemic” (January 2021-December 2022) overall and across subgroups based on rurality and social vulnerability index.

Methods:

The study included all Medicare fee-for-service (FFS) and Medicare Advantage (MA) beneficiaries with CV risk factors (hypertension, diabetes mellitus, or hyperlipidemia) or CVD (coronary heart disease, heart failure, atrial fibrillation, stroke) between January 2018 and December 2022. A Poisson-lognormal regression model was fit to compare the adjusted incidence rate ratio of acute care visits, outpatient visits, and all-cause mortality during the late pandemic (January 2021-July 2022) compared with the pre-pandemic period (January 2018-December 2019).

Results:

There were 51,355,577 beneficiaries in the pre-pandemic period and 55,214,638 in the late pandemic period with CV risk factors or CVD. Hospital visit rates were lower in the late pandemic period compared with the pre-pandemic period (adjusted incidence rate ratio [aIRR] 0.918 [95% CI 0.915, 0.922]), and declines were evident among rural and urban beneficiaries, as well as across all levels of social vulnerability and in both fee-for-service and Medicare Advantage. In contrast, use of outpatient visits was higher during the late pandemic period (aIRR 1.141 [1.134, 1.149]); these increases were on average higher among urban communities and Medicare Advantage beneficiaries. Mortality was higher during the late pandemic period (aIRR 1.248 [1.240, 1.257]), and surges in mortality among this population tracked with national COVID-19 mortality. Beneficiaries in the most socially vulnerable communities (aIRR 1.116 (1.106, 1.124)) and those covered by Medicare Advantage (aIRR 1.342 [1.329, 1.356]) experiencing the greatest increase.

Conclusions:

In the later stages of the pandemic, among Medicare beneficiaries with CV risk factors or established CVD, hospitalizations were lower than pre-pandemic, outpatient care was used more frequently, and mortality was significantly elevated across risk groups and geographies, with the greatest increases seen in vulnerable communities and those covered by Medicare Advantage.

Condensed Abstract:

We sought to examine utilization of healthcare services and mortality rates in the late pandemic period (2021–2022) compared to the pre-pandemic period (2018–2019) among Medicare beneficiaries with cardiovascular risk factors or disease. We found that hospital visit rates were lower, use of outpatient visits was higher, and mortality was higher in the late pandemic period compared with pre-pandemic. In sum, there were persistent differences in care utilization and clinical outcomes in the late pandemic period compared to pre-pandemic among individuals with cardiovascular risk factors or disease.

Introduction

The early phase of the COVID-19 pandemic led to tremendous morbidity and mortality as well as enormous disruptions in inpatient and outpatient care that were most pronounced among socially vulnerable populations. Older adults with cardiovascular risk factors and/or established cardiovascular disease may have been particularly susceptible both to COVID itself and to these care disruptions, as evidenced by precipitous declines in myocardial infarction and stroke hospitalizations,^1–4 and associated increases in population-level cardiovascular mortality during the first few months of the pandemic.^5–8 However, little is known about whether these patterns persisted into the later phases of the pandemic.

Understanding changes in healthcare utilization and outcomes for older adults with CV risk factors and/or disease is critically important and could have major downstream implications for health care spending, payment policy, and population health. First, understanding the impact of COVID on older adults with cardiovascular disease could have implications for future COVID waves or indeed for future viral pathogens or pandemics more broadly. Second, understanding changes in care utilization is also important. Disruptions in important health care services, including preventive screenings, treatment of risk factors like diabetes and hypertension, and procedural care for cardiovascular conditions (e.g., ischemic heart disease, stroke), could have led to a surge in pent up hospitalizations following the early pandemic.^9,10 On the other hand, patients could have continued to avoid seeking care as the pandemic went on due to fear of contracting the virus, were lost to follow-up, or experienced reduced access to care.^11–13 Strains on hospital capacity, increases in out of hospital deaths whether due to COVID or otherwise, or shifts in coding or admission practices could have also impacted patterns of care. In addition, understanding changes in health care services and clinical outcomes among socially vulnerable populations, who disproportionately experienced both the direct clinical impacts of COVID and health care disruptions,^14–18 could help inform public health and policy strategies to improve cardiovascular health in the years that follow the pandemic.

Therefore, in this national study, we aimed to answer 3 questions. First, how did health care utilization among Medicare beneficiaries with CV risk factors and/or established CVD in the late pandemic period compare to pre-pandemic levels? Second, how did mortality rates compare between the late pandemic period and pre-pandemic levels? And finally, how did any observed changes vary by rurality, social vulnerability, or fee-for-service versus Medicare Advantage coverage?

Methods

Data Sources and Study Population

We used the Virtual Research Data Center (VRDC) to perform a longitudinal study of the Medicare population from January 1, 2018, through December 31, 2022. The VRDC captures 100% of beneficiary claims across all hospital-based (inpatient, observation, emergency department) and outpatient care settings, including both in-person and telehealth visits. We defined the pre-pandemic period as January 1, 2018 through December 31, 2019 and the late pandemic period as January 1, 2021 through December 31, 2022. We also obtained data on weekly national COVID-19 death counts per 100,000 population from the Centers for Disease Control and Prevention,¹⁹ and approximate start and end dates for major COVID variant waves from the New York Times COVID variant tracker.²⁰

Our study population included all fee-for-service and Medicare Advantage beneficiaries with cardiovascular risk factors (hypertension, diabetes mellitus, or hyperlipidemia) and/or established cardiovascular disease (coronary heart disease, heart failure, atrial fibrillation, or stroke). Exclusions are shown in eFigure 1. We followed Medicare’s Chronic Conditions Warehouse (CCW) methodology to identify each qualifying condition. This algorithm uses diagnosis codes in claims during the preceding 1–2 years to identify beneficiaries with these common clinical conditions, and is typically applied on an every-6-month or annual basis.²¹ We adapted these algorithms to capture beneficiaries with CV risk factors or CVD on a rolling basis, thus reducing ascertainment bias associated with end-of-year comorbidity reporting and aligning at-risk periods more precisely by allowing entry into or exit from the cohort at any point in the study period. For our purposes, entrance into the surveillance cohort begins one day after the claim that qualifies a beneficiary as having a chronic condition (which will exclude the initial qualifying utilization event from our outcomes), and exit from the surveillance cohort occurs one day after the one or two year period specified by CCW for each chronic condition, unless a beneficiary has an additional qualifying event that extends their membership in the at risk cohort. We used the VRDC to determine characteristics of each beneficiary, including age, gender, race and ethnicity, and dual enrollment in Medicaid. For race and ethnicity, we used the RTI variable, which augments enrollment data with additional information based on surname and residence.²² For subgroup analyses, in addition to FFS versus MA coverage, we assessed rurality and social vulnerability. We assessed rurality using Core Based Statistical Areas based on county level FIPS codes, to determine whether beneficiaries lived in metropolitan or micropolitan areas, which are explicitly defined, or rural areas, which are undefined.²³ Beneficiaries’ FIPS code of residence were also used to determine the social vulnerability index (SVI) of their communities. The SVI was developed by the Centers for Disease Control and Prevention and combines US Census Bureau data on 15 variables across 4 domains: socioeconomic status, housing composition and disability, minority status and language, and housing type and transport.^24,25 The overall SVI provides a county-level percentile rank from lowest to highest vulnerability, with higher percentile ranks indicating higher levels of social vulnerability. We categorized beneficiaries into quartiles based on their national SVI ranking.

Outcomes

We evaluated hospital visits (inpatient, observation, and emergency room stays) and outpatient visits (in-person and telehealth) per beneficiary-year. In addition, we also examined changes in all-cause mortality, both overall and in-hospital (defined as death with an acute care encounter within +/− one day).

Statistical Analysis

We first compared beneficiary sociodemographic characteristics during the pre- and late pandemic periods. We then determined hospital visits, outpatient visits, and all-cause mortality, as well as incidence rates for each outcome per beneficiary-year, and plotted raw data. To the overall mortality graph, we added a trace line representing the number of deaths from COVID per 100,000 population each month during the pandemic, as well as markers for each variant wave, as described above.¹⁹

A hierarchical Bayesian Poisson-lognormal regression model was fit to outcome counts during the late pandemic (1/2021–12/2022) compared with the pre-pandemic onset period (1/2018–12/2019),²⁶ adjusted for age, sex, race, and quarter-year with an offset term for beneficiary-years. We excluded the year 2020 from the model to focus on the long-term average impacts of the pandemic on the health care system rather than the acute disruptions, which have been documented in prior work,^9,27 and additionally exclude at risk time and outcomes for MA beneficiaries from 7/2022 – 12/2022 to accommodate delayed data delivery schedules for Medicare Advantage Encounter data where longer hospitalizations may affect our cohort ascertainment and outcomes quantification. Our model also included a state-level random effect as well as a variable indicating whether a beneficiary had a billing code indicating acute COVID-19 infection in the previous 90 days, ascertained similarly to our CCW definitions described above. Full details of our modeling approach are provided in the supplemental appendix. We then repeated our main analysis across two key subgroupings, rurality (metropolitan, micropolitan, rural), and social vulnerability (quartiles), and additionally, present results from our main analysis by enrollment in FFS versus MA. The FFS versus MA analysis was added post-hoc and should be considered exploratory.

In all models, we carefully specified relatively non-informative priors for all parameters, and assessed posterior convergence diagnostic criteria, such as Rhat and traceplots for all parameters of interest within the VRDC environment, as is required by our data use agreement. We observed no evidence of model fit, convergence, or posterior chain mixing issues. Full details of our modeling specification, prior parameterization, and explicit diagnostic assessment procedures with criteria are provided in the supplemental appendix for researchers interested in replicating our approach.

This study was approved by the Office of Human Research Protection at Washington University in St. Louis, and the requirement for informed consent was waived due to the de-identified nature of the data. Analyses were performed using SAS Enterprise Guide 8.3, R version 4.4.1,²⁸ and Stan version 2.32.2.²⁹ P-values less than 0.001 were considered statistically significant.

Results

There were 51,355,577 and 55,214,638 unique Medicare beneficiaries with cardiovascular risk factors and/or disease in in the 2018–2019 and 2021–2022 time periods respectively, contributing a total of 206,781,871 beneficiary-years of data (Table 1). Beneficiary characteristics were generally similar between the pre- and late pandemic periods, though in the later period beneficiaries were slightly older, less often white (75.0% versus 73.9%), more often urban (82.1% versus 82.5%), and more likely to be enrolled in Medicare Advantage (37.6% versus 46.7%).

Table 1.

Characteristics of Medicare Beneficiaries with Cardiovascular Risk Factors and/or Disease

	Pre-Pandemic	Late Pandemic
	January 2018 to December 2019	January 2021 to December 2022
	(N=51,355,577)	(N=55,214,638)
Age at start month
<65	13.5	11.8
65 – 74	49.5	51.2
75 – 84	25.9	26.6
85+	11.1	10.4
Sex
Male	44.4	44.5
Female	55.6	55.5
Race/Ethnicity a
White	75.0	73.9
Black	10.8	10.8
Hispanic	8.1	8.5
Other	6.1	6.7
Medicare Enrollment Type
Fee-for-service	62.4	53.3
Medicare Advantage	37.6	46.7
Rurality b
Urban	82.1	82.5
Micropolitan	10.1	10.0
Rural	7.8	7.6
Social Vulnerability Index c
Quartile 1 (least vulnerable)	13.9	14.2
Quartile 2	24.6	24.7
Quartile 3	29.2	29.1
Quartile 4 (most vulnerable)	32.3	32.0
Chronic Conditions
Acute Myocardial Infarction	3.3	3.4
Atrial Fibrillation	17.5	17.7
Diabetes	37.4	36.9
Hyperlipidemia	82.5	84.8
Hypertension	83.5	82.6
Ischemic heart disease	29.3	29.1
Stroke/TIA	13.8	12.8

Percentages of beneficiary-months are shown unless otherwise specified. For chronic conditions, only cardiovascular risk factors and/or cardiovascular diseases are shown.

^aRace and ethnicity was determined based on RTI categories

^bRurality was defined based on Core-Based Statistical Areas

^bSocial vulnerability index of beneficiaries’ place of residence was categorized in national quartiles

Hospital Visits

Unadjusted trends in hospital visits per beneficiary-year are shown in Figure 1, both overall and across key subgroups. Hospital visit rates were higher among rural beneficiaries compared with their urban counterparts over the entire study period. In addition, beneficiaries residing in communities with high social vulnerability (quartile 4) had markedly higher hospital visit rates than those in the lowest (quartile 1). Rural beneficiaries had higher hospital visit rates than their urban counterparts. Patterns were similar when these visits were disaggregated into ED-only, observation, and inpatient stays (eFigures 2–4).

Figure 1. Hospital Visits Among Medicare Beneficiaries with Cardiovascular Risk Factors and/or Disease, January 2018 to June 2022.

The figure shows observed trends in the number of hospital visits per 30 beneficiary-days, defined as an inpatient, observation, or emergency room visit, among Medicare beneficiaries with cardiovascular risk factors or established cardiovascular disease from January 2018 through July 2022, overall (bottom right) and by rurality (top left), social vulnerability index (top right), and fee-for-service versus Medicare Advantage enrollment (bottom left). Shading reflects our study pre-pandemic period (2018–2019), early COVID (2020), and late pandemic (2021–2022).

Model results suggested that, on average, hospital visit rates were statistically significantly lower in the late pandemic period compared with the pre-pandemic period (adjusted incidence rate ratio [aIRR] 0.918 [95% CI 0.915, 0.922]) (Table 2). Declines in inpatient care were similar across urban (aIRR 0.895 [0.890, 0.900]), micropolitan (aIRR 0.873 [0.867, 0.884]), and rural (aIRR 0.877 [0.870, 0.884]) areas. These patterns were also evident across levels of SVI, with a lower drop among the least vulnerable (aIRR 0.895 [0.889, 0.901]) than the most vulnerable (aIRR 0.875 [0.870, 0.880)], p for interaction <0.001) beneficiaries. Fee-for-service beneficiaries experienced steeper declines in hospital visits on average (aIRR 0.885 [0.881, 0.890]) than those enrolled in Medicare Advantage (aIRR 0.967 [0.962, 0.973], p for interaction<0.001). Findings were similar when these visits were disaggregated into ED-only, observation, and inpatient stays (eTable 1).

Table 2.

Health Care Utilization Among Medicare Beneficiaries with Cardiovascular Risk Factors and/or Disease

		Incidence Rate Ratio (95% CI) Late Pandemic vs. PrePandemic	P value for Pre- vs Late-Pandemic	P value for interaction
Hospital Visits
Overall		0.918, (0.915, 0.922)	<0.001
Rurality	Urban	0.895, (0.890, 0.900)	<0.001	Ref
	Micropolitan	0.873, (0.867, 0.879)	<0.001	0.92
	Rural	0.877, (0.870, 0.884)	<0.001	0.035
SVI	Quartile 1 (least vulnerable)	0.895, (0.889, 0.901)	<0.001
	Quartile 2	0.888, (0.883, 0.893)	<0.001	0.56
	Quartile 3	0.879, (0.874, 0.884)	<0.001	<0.001
	Quartile 4 (most vulnerable)	0.875, (0.870, 0.880)	<0.001	<0.001
Enrollment	Fee-for-Service	0.885, (0.881, 0.890)	<0.001
	Medicare Advantage	0.967, (0.962, 0.973)	<0.001	<0.001
Outpatient Visits
Overall		1.141, (1.134, 1.149)	<0.001
Rurality	Urban	1.090, (1.084, 1.096)	<0.001
	Micropolitan	1.025, (1.018, 1.031)	<0.001	<0.001
	Rural	1.004, (0.997, 1.011)	0.286	<0.001
SVI	Quartile 1 (least vulnerable)	1.054, (1.047, 1.060)	<0.001
	Quartile 2	1.053, (1.046, 1.059)	<0.001	0.042
	Quartile 3	1.049, (1.043, 1.055)	<0.001	<0.001
	Quartile 4 (most vulnerable)	1.057, (1.050, 1.063)	<0.001	<0.001
Enrollment	Fee-for-Service	1.093, (1.084, 1.102)	<0.001
	Medicare Advantage	1.215, (1.204, 1.225)	<0.001	<0.001
Mortality
Overall		1.248, (1.240, 1.257)	<0.001
Rurality	Urban	1.076, (1.069, 1.083)	<0.001
	Micropolitan	1.078, (1.068, 1.089)	<0.001	0.083
	Rural	1.089, (1.077, 1.100)	<0.001	0.031
SVI	Quartile 1 (least vulnerable)	1.035, (1.024, 1.046)	<0.001
	Quartile 2	1.056, (1.048, 1.064)	<0.001	0.0017
	Quartile 3	1.080, (1.072, 1.088)	<0.001	<0.001
	Quartile 4 (most vulnerable)	1.116, (1.106, 1.124)	<0.001	<0.001
Enrollment	Fee-for-Service	1.187, (1.176, 1.197)	<0.001
	Medicare Advantage	1.342, (1.329, 1.356)	<0.001	<0.001

Abbreviations: SVI=social vulnerability index

Outpatient Visits

Unadjusted trends in outpatient visit rates (in-person and telehealth combined) are shown in Figure 2. Differences between subgroups were generally similar to hospital visits, with the exception of rural beneficiaries, who had lower outpatient visit rates (but higher hospital visit rates) than their urban counterparts. When these visits were disaggregated into in-person and telehealth visits, it was evident that the majority of the growth in outpatient visits was due to an increase in telehealth (eFigures 5–6)

Figure 2. Outpatient Visits Among Medicare Beneficiaries with Cardiovascular Risk Factors and/or Disease, January 2018 to June 2022.

The figure shows observed trends in the number of outpatient visits per 30 beneficiary-days, defined as an in-person or telehealth evaluation and management visit, among Medicare beneficiaries with cardiovascular risk factors or established cardiovascular disease from January 2018 through July 2022, overall (bottom right) and by rurality (top left), social vulnerability index (top right), and fee-for-service versus Medicare Advantage enrollment (bottom left). Shading reflects our study pre-pandemic period (2018–2019), early COVID (2020), and late pandemic (2021–2022).

Model results suggested that outpatient visits (combined in-person and telehealth) were higher during the late pandemic period than in the pre-pandemic period (aIRR 1.141 [1.134, 1.149]) (Table 2). Increases in outpatient visits were greater among beneficiaries in urban areas (aIRR 1.090 [1.084, 1.096]), than among those residing in micropolitan (1.025 [1.018, 1.031], p for interaction <0.001) or rural areas (1.004 ([0.997, 1.011], p for interaction <0.001). Increases were qualitatively similar among all levels of SVI (1.054, 1.053, 1.049, 1.057 for least to most vulnerable). The rise in outpatient visits was more pronounced among Medicare Advantage beneficiaries (aIRR 1.215 [1.204, 1.225]) than those in fee-for-service (aIRR 1.093 [1.084, 1.102], p for interaction <0.001).

All-Cause Mortality

Unadjusted trends in mortality are shown in Figure 3, and were higher among rural compared with urban beneficiaries, and among the most versus least socially vulnerable beneficiaries. Peaks and troughs in mortality tracked largely with national COVID mortality. Model results suggested that mortality was higher during the late pandemic period than in the pre-pandemic period (aIRR 1.248 [1.240, 1.257]) (Table 2). Increases in mortality were similar in magnitude for urban (aIRR 1.076 [1.069, 1.083]), micropolitan (aIRR 1.078 [1.068, 1.089]) and rural (aIRR 1.089 (1.077, 1.100)]) beneficiaries. Increases in mortality were larger in the more socially vulnerable areas (1.035, 1.056, 1.080, 1.116 for SVI quartiles from least to most vulnerable, p<0.001 for q3 and q4 versus q1). Increases in all-cause mortality were higher in Medicare Advantage (aIRR 1.342 [1.329, 1.356]) than fee-for-service (aIRR 1.187 [1.176, 1.197], p for interaction <0.001). When we analyzed the subset of only in-hospital mortality, the increases were slightly smaller (aIRR 1.136 [1.127, 1.145]) overall and among key subgroups (eTable 2), though the overall patterns were similar, with no difference in the increase in mortality by rurality, a higher increase among more vulnerable areas, and a higher increase in Medicare Advantage.

Figure 3. Mortality Among Medicare Beneficiaries with Cardiovascular Risk Factors and/or Disease, January 2018 to June 2022.

The figure shows observed trends in mortality per 30 beneficiary-days, among Medicare beneficiaries with cardiovascular risk factors or established cardiovascular disease from January 2018 through July 2022, overall (bottom right) and by rurality (top left), social vulnerability index (top right), and fee-for-service versus Medicare Advantage enrollment (bottom left). Shading reflects our study pre-pandemic period (2018–2019), early COVID (2020), and late pandemic (2021–2022). The dotted line represents national counts of COVID deaths as reported by the Centers for Disease Control and Prevention. Greek characters represent different dominant variants of COVID.

Discussion

In this national study of all Medicare beneficiaries with documented cardiovascular risk factors and/or established cardiovascular disease, we found that, on average, hospital visits were lower in 2021 and 2022 than pre-pandemic. In contrast, outpatient visit rates increased beyond levels seen prior to the pandemic, driven largely by a rise in telehealth visits. Mortality rates were significantly higher among beneficiaries with cardiovascular risk factors or conditions in 2021 and 2022 than pre-pandemic.

Our first finding was an increase in outpatient visits, particularly telehealth, and a decrease in hospitalizations. While this observational study does not allow causal inference, there are a number of potentially contributing factors to the patterns we observed. One possibility is that the pandemic drove durable changes in how people utilize health care; the increase in telehealth has been broadly documented,³⁰ and while debate remains about whether telehealth care is a substitute for in-person care or an add-on, its utilization is likely here to stay. Indeed, greater use of outpatient care, which has traditionally been considered underutilized - particularly for high-risk groups - and a higher threshold to seek hospital care are often seen as optimal use patterns. Lower use of hospital care could also be due to a healthy survivor bias, if the post-COVID-wave survivor cohorts were healthier and thus demonstrated less care-seeking behavior.

However, the fact that we found these patterns paradoxically coupled with higher mortality raises concerns, though also has myriad potential explanations. Some of the increased mortality was certainly due to acute COVID infection; though we controlled for the presence of COVID diagnosis codes in our models, the waves of mortality evident in the raw mortality graphs emphasize that COVID, much like flu before it, likely drove seasonal variations in disease exacerbation and mortality among patients with cardiovascular risk factors or disease.³¹ Indeed, peaks in mortality closely mirrored the timing of major COVID-19 waves-including Spring 2020, Winter 2020–21, Fall 2021 (Delta), and Winter 2021–22 (Omicron)-suggesting that the observed increases in mortality during the late pandemic period were largely driven by these infection surges. Some of the elevated mortality may also be driven by other after-effects of the virus rather than its acute impact. For example, COVID-19 leads to higher levels of inflammation, which contribute to worse overall clinical cardiovascular outcomes both over the short- and long-term.^32–34 In addition, social and behavior risk factors (e.g., poverty, dietary patterns, psychosocial stress) that are strongly linked with cardiometabolic health worsened during the initial phase of the pandemic.^35–39

System-level effects could also contribute, though these are difficult to study given data limitations. Hospital capacity challenges, including staff shortages, care delivery challenges, or constrained resources, could have impacted outcomes in the acute or chronic setting, both for COVID and for non-COVID illnesses, particularly during major spikes in COVID activity. Changes in coding or admission practices could confound our findings. Patient factors could also contribute; prior studies showed that people deferred care for heart attacks and strokes during the early days of the pandemic, and often arrived at the hospital with more serious, complicated versions of their clinical syndrome than would otherwise have occurred.^12,40,41 An increase in out-of-hospital deaths could also contribute; the increase we found in all-cause mortality was greater than the increase we found in in-hospital mortality, though such evidence is circumstantial at best. Together, both direct and indirect effects of the pandemic may have contributed to the persistently higher mortality rates that we observed through the end of 2022.

The differences we found by geography and by social risk group are also important. For example, urban beneficiaries experienced a significant rise in outpatient visits driven by telehealth use, while rural beneficiaries – who already faced barriers accessing outpatient cardiovascular care⁴² – experienced no change. People living in the most socially vulnerable counties experienced the largest increase in all-cause mortality, and the widening of disparities in the late pandemic period suggest that there is great need for improvements in health care access and infrastructure in vulnerable communities. These communities, which were hardest-hit in the early waves of COVID both in terms of health and economic strain,⁴³ may have lost key health infrastructure or access points that enabled use of non-emergent health care services for COVID or other illnesses. Aid given to businesses and health care organizations during the pandemic often disproportionately benefited those with the infrastructure to easily claim it, potentially worsening structural and systemic inequities in our health care system.⁴⁴ Overall, our findings support the urgent need for targeted health system, community, and public health to prevent geographic disparities in outpatient care and cardiovascular health from widening further.

Finally, we also observed major differences across Medicare plan type. Medicare Advantage beneficiaries experienced lesser declines in hospital visits, but a greater increase in outpatient visits, and a greater increase in mortality, than those enrolled in traditional fee-for-service Medicare. During the study period, beneficiaries were steadily shifting into MA, and the proportion of beneficiaries switching back and forth between MA and FFS was also growing;⁴⁵ while our analyses controlled for basic demographics and comorbidities, as well as area of residence, it is possible that these patterns reflect residual confounding related to the population joining MA. Given that this analysis was post-hoc, it should be considered exploratory and future studies should evaluate these changes with more clinically-rich data.

Our findings build upon one recent analysis that found that ED visits were lower but rates of admission from the ED higher for non-COVID-related conditions after the onset of the pandemic.⁴⁶ We evaluate a comprehensive measure of overall acute care utilization (inpatient stays, ED visits, observation stays) in a higher-risk population – those with cardiovascular risk factors or disease – who could potentially be more susceptible to the effects of the pandemic. Additionally, our study provides new insights on changes in outpatient visits as well as a concerning sustained increase in all-cause mortality among people with cardiovascular risk factors and/or disease through 2022. Together, these findings build upon early studies highlighting the direct and indirect tolls of the pandemic,^8,38,47–52 by providing a comprehensive evaluation of longer-term changes in health care utilization and mortality in the entire Medicare population, both overall and by geographic and social vulnerability.

There are limitations to our study. First, Medicare claims data do not include granular clinical information, and as such we are unable to measure patients’ blood pressure, diabetes control, or other biometric or laboratory values. Second, we studied patients with cardiovascular risk factors or overt cardiovascular disease in order to focus on a population with defined care needs, but our findings may not apply to people without these conditions, and we did not explicitly account for clinical or claims based indicators, such as medication adherence patterns, non-CV chronic conditions, or complex utilization measures like frailty in our analysis, which may serve to further explain changes in utilization and mortality patterns pre-to-post pandemic. Third, associations seen in observational data do not imply causality, and further work is needed to disentangle the multiple potential mechanisms we outlined above. Our analytic choices also matter; this analysis compares marginal averages before and during the late pandemic period, which emphasizes persistent differences in care and outcomes, and allows the flexibility to investigate differential changes in levels by subgroupings of interest, such as rurality, enrollment choice, and measures describing areal social vulnerability. In contrast, an interrupted time series (ITS) design approach using an ITS design may have revealed additional collapse-and-rebound dynamics at the cost of flexibility to investigate multiple differential changes by subgroup. A strength of this analysis is that it included all Medicare beneficiaries with CV risk factors and/or disease, including those enrolled in fee-for-service or Medicare Advantage; this allowed us essentially complete capture of hard endpoints like hospitalization and mortality for the older adult US population across a five-year period.

Conclusion

In conclusion, we found that for patients with cardiovascular risk factors or established cardiovascular disease, hospitalizations remained lower in the late pandemic period than pre-pandemic, outpatient care was used more frequently, and most importantly, mortality remained significantly elevated across risk groups and geographies, with the greatest increases seen in the most vulnerable communities.

Abbreviations:

COVID-19

Coronavirus disease 2019

CV

cardiovascular

CVD

cardiovascular disease

ED

emergency department

FFS

Fee-for-Service

MA

Medicare Advantage