This cross-sectional study used data from multiple sources. The population of interest was Medicare fee-for-service (FFS) beneficiaries aged ≥65 years. The primary data source is 2019 administrative claims data from the Centers for Medicare and Medicaid Services for Medicare Part A and Part B beneficiaries aged ≥65 years enrolled in FFS plans in 3,142 US counties.¹⁶ Data from the 2019 American Community Survey¹⁷ were used to calculate county-level population characteristics. Population counts at the census tract, county, and state levels were obtained from the 2020 US Census.¹⁸ Data from the 2020 Broadband Deployment Report by the Federal Communications Commission (FCC)¹⁹ were obtained to measure county-level broadband availability. This study was considered exempt from Institutional Review Board review under federal regulations.

CR demand was defined at the county level as the number of unique Medicare FFS beneficiaries aged ≥65 years who had a CR-qualifying event in 2019 (N=515,712) based on county of residence. To estimate CR demand at the census tract level, county-level CR demand was distributed across census tracts within each county proportionally, based on the population aged ≥65. Further details about the analytic cohort and CR-qualifying events are included in the Supplemental Material; methods were consistent with prior surveillance.⁶

The CR supply for each facility was defined as the total number of unique Medicare FFS beneficiaries aged ≥65 years who received at least one CR service in 2019 (N=178,287), based on the assumption that most facilities operate at or near capacity. CR services were identified using Healthcare Common Procedure Coding System (HCPCS) codes for standard (93797, 93798) and intensive CR (G0422, G0423). CR facilities were geocoded using the street address of the facility.

CR demand met was estimated using spatially adaptive floating catchments (SAFCs).²¹ An SAFC is a collection of geographic areas centered around a facility that expands outward until the estimated demand for services equals or slightly exceeds the number of services provided at a facility. We calculated SAFCs for every CR facility in the United States. A graphical explanation of SAFCs is included in the Supplement (Figures S1 and S2). A county’s “CR demand met” was the sum of CR demand in the census tracts that were covered by SAFCs.

CR availability for each county was measured as the percent of total CR demand that was met (100 * CR demand met / total demand for CR). Counties were classified based upon their CR availability: (1) lowest CR availability (percent demand met <35%), (2) medium CR availability (demand met from ≥35% to <70%), and (3) highest CR availability (demand met ≥70%). Counties with the lowest CR availability were also referred to as CR deserts.

All measures for county-level CR availability generated by this study will be available at https://chronicdata.cdc.gov/.

Census block-level broadband data was aggregated to the county level. County-level broadband availability was defined as the percent of the population in a county with access to ≥1 internet provider of residential nonsatellite 100Mbps download and 10Mbps upload. Based on previous literature, this level of service could properly support hybrid CR.²² Satellite internet was not considered to be sufficient for hybrid CR because its connection is less consistent than wired broadband and it is often more expensive than wired broadband.²³

CR deserts (counties with <35% CR availability) were further stratified into CR deserts with majority-available broadband (≥50% broadband availability) vs dual deserts (<50% broadband availability).²²

County-level sociodemographic characteristics obtained from the American Community Survey included race/ethnicity (American Indian or Alaska Native, Asian or Pacific Islander, non-Hispanic Black, Hispanic, non-Hispanic White), age (<35, 35–54, 55–64, ≥65), sex (male, female), health insurance status (% insured), education (% no college diploma), median income, poverty rate (<100% of the federal poverty threshold), and eligibility for the Affordable Connectivity Program (<200% of the federal poverty threshold). Urban/rural status was classified based on the National Center for Health Statistics’ Urban-Rural Classification Scheme (urban = large central metro, large fringe metro, medium metro, or small metro; rural = micropolitan or noncore).²⁴

ArcGIS 10.6.1 was used to calculate the location of each CR facility along with the center of population density for each census tract. A Monte Carlo simulation was used to estimate demand met in each county from 1,000 randomized iterations of SAFC calculations.²⁵ A Monte Carlo simulation was necessary because SAFC analyses required that CR demand be calculated at the census tract level, but Medicare FFS data for CR-qualifying events were available only at the county level. Therefore, county-level demand was disaggregated to the census tract level using a multinomial distribution, weighted by the size of the population ≥65 years in each census tract, to calculate each iteration of the demand met. The median demand met of all iterations for each county was divided by the corresponding county’s demand to calculate a median percent demand met.

Trends were compared in county-level population characteristics (race/ethnicity, sex, age, insurance, education, income) by CR availability (highest, medium, lowest) using an empirical Monte Carlo Cuzick test for trend.^26,27 The Supplemental Material contains further information on the statistical test used. A chi-square trend test was used for urban-rural status. Differences in county-level population characteristics—for CR deserts with majority-available broadband and dual deserts—were compared using an unpaired t-test weighted by county population, or a chi-square test (for urban-rural status). Statistical analyses were conducted in R 4.2.1, including software packages Rcpp and RcppArmadillo to assist with computationally intensive calculations.^28–33

In our cohort derivation, we identified 516,668 Medicare FFS beneficiaries aged ≥65 years with a CR qualifying event in 2019, continuously enrolled in Medicare parts A and B for 1 year. The percent of inpatient CR-eligible Medicare beneficiaries was 79.0% and the percent of outpatient CR-eligible Medicare beneficiaries was 21.0%. 52.7 percent of beneficiaries had an AMI, followed by PCI (44.6%), heart valve repair or replacement (12.62%), CABG (10.32%), and heart or heart-lung transplant (0.1%). We identified a national CR capacity of 178,287 beneficiaries.

There were 3,142 US counties in 2020, representing a total adult population of nearly 247 million (Table 1). Of all US counties, 1,691 (53.8%) had the lowest levels of CR availability (CR deserts), 889 (28.3%) had medium CR availability, and 562 (17.9%) had the highest levels of CR availability. Among the total adult population, 116 million adults (47.2%) lived in CR desert counties.

As counties had decreasing CR availability, counties had fewer non-Hispanic White residents and more Hispanic residents, along with an increased percentage of residents without a college diploma (all p<0.01). Though no patterns were found in poverty rates across CR availability categories (p=0.39), CR deserts had statistically higher percentages of people eligible for the Affordable Connectivity Program (p<0.01).

Among CR desert counties, 77.4% (N=1,308) of counties (comprising ≈113 million people) had ≥50% broadband availability (Table 2). The remaining 22.6% of the CR desert counties (N=383, population = 3.6 million people) were dual deserts.

There were statistically significant differences in the sociodemographic composition of CR desert counties with majority-available broadband compared to dual deserts (Table 2). Dual desert counties had statistically significantly higher percentages of American Indian or Alaska Native (weighted mean: 5.3% vs 0.6%) and non-Hispanic White residents (67.9% vs 55.8%), and lower percentages of Asian or Pacific Islander (0.7% vs 6.3%) and Hispanic residents (9.6% vs 17.3%) (all p<0.01). Dual deserts had ‘less favorable’ socioeconomic conditions, including lower median income ($48,878 vs $69,576), higher poverty levels (18.2% vs 12.8%), and greater eligibility for the Affordable Connectivity Program (34.8% vs 25.8%) (all p<0.01). Compared to urban counties, a greater percentage of rural counties were dual deserts (84%, n=321).

There were strong geographic patterns in CR and broadband availability (Figure 1). CR desert counties (with and without broadband deserts) were located throughout the Northeast, South and West. Dual CR-broadband desert counties were concentrated primarily in parts of Appalachia, parts of the Deep South, parts of the Southwest, and Alaska. Counties with the highest CR availability were located predominantly in parts of the Midwest and Great Plains.

By state, CR availability varied widely (Figure 2). States with the highest percentage of the adult population living in CR desert counties were Hawaii (100%), the District of Columbia (100%), Nevada (96.8%), New York (92.1%), and New Mexico (91.3%). States with the lowest percentage of the adult population living in CR desert counties included New Hampshire (3.2%), Minnesota (3.7%), Nebraska (4.2%), Wisconsin (6.2%), and Wyoming (6.3%).

Across all states, the percentage of the adult population living in a dual desert ranged from 0% to 12.4%. Thirteen states had approximately 0% of their population living in a dual desert. States with the highest percentage of the population living in a dual desert were Mississippi (12.4%), Alaska (12.1%), New Mexico (9.1%), Montana (9.0%), and Arkansas (8.9%).

This study had 2 key strengths. The data used in this study comprised robust national datasets that allowed us to estimate CR availability at the county level. Using sophisticated methodology, the size of CR facilities and the surrounding areas they serviced was accounted for. This methodology also allowed for CR facilities located close to county or state borders to contribute to the demand met in multiple administrative regions.²¹ This study was also subject to limitations. First, the estimates of CR demand and supply were based on the Medicare FFS population, limiting the generalizability of these findings to younger individuals. Data about CR demand at the census tract level and data about supply measured at the individual level would more precisely measure CR availability.²¹ Demand for CR for the Medicare FFS population was unknown at the census tract level and had to be randomly allocated to each census tract to obtain median demand met estimates. Our estimates of facility-level CR capacity may be over- or under-estimated. By basing CR capacity on facility-level persons served, we assume that clinics are operating at full capacity, which could under-estimate capacity if facilities have unused capacity or the ability to expand. Alternatively, over-estimates are possible because supply for CR services considered only the number of unique beneficiaries billed, regardless of the number of services these beneficiaries were billed for. Third, our estimates of CR availability using 2019 data are likely higher than current CR availability since CR availability and participation declined during COVID-19 and have not returned to prepandemic levels.⁸

Future studies could examine CR availability in populations other than Medicare enrollees, or differences in CR availability patterns between urban and rural counties. The potential role of asynchronous CR, which could be conducted using simple telephones and smartphones, could also be explored. Lastly, additional information could be revealed by (1) studies utilizing more granular cardiovascular disease data and (2) datasets that connect individuals who qualify for CR to billed CR sessions or that include a broader population of qualified patients.