J. Mick Tilford, PhD

To quantify geographic variation in home health expenditures per Medicare home health beneficiary and investigate factors associated with this variation.

 Retrospective study design analyzing US counties in which at least 1 home health agency served 11 or more beneficiaries in 2016. Several sources of 2016 national public data were used.

 The key variable is county-level Medicare home health expenditures per home health beneficiary. Counties were grouped into quintiles based on per-beneficiary expenditures. Analyses included calculation of coefficients of variation, computation of the ratio of 90th percentile to 10th percentile in expenditures, and linear regression predicting expenditure. The control variables included characteristics of patients, agencies, and communities.

Significant variation in home health expenditures was identified across county quintiles, with a 90th-to-10th-percentile expenditure ratio of 2.5. The percentage of for-profit agencies in the lowest quintile was 15.7 compared with 81.7 in the highest quintile of spending. Unadjusted spending differed by $3864 (95% CI, $3793-$3936), compared with $3611 (95% CI, $3514-$3708) in the adjusted model, between counties in spending quintiles 1 and 5. Although state fixed effects explained nearly 20% of the variation in home health expenditures, 42% of the variation remained unexplained.

 Home health care exhibits considerable unwarranted variation in per-patient expenditures across counties, signifying inefficiency and waste. Given the expected growth in home health demand, strategies to reduce unwarranted geographic variation are needed.

Existing research on home health expenditures uses home health data more than 2 decades old. US home health expenditures rose by 113% between 2000 and 2016, from $8.5 billion to $18.1 billion. The Medicare program has implemented several policies in the past decade to combat the growth in expenditures. This study finds the following:

A 2.5-fold spending difference per beneficiary exists between the 90th and 10th spending percentiles.

Patient characteristics explained only 16% of the variation in home health expenditures.

More than 40% of spending variation remained unexplained in the study, suggesting substantial inefficiency and waste in the home health care program.

Unwarranted variation in the utilization of health services, or variation not related to differences in patient needs or conditions, is pervasive across health care settings in the United States.

 Unwarranted variation in services typically leads to increased health care spending without a concomitant improvement in health care outcomes.

 Reducing unwarranted variation in health care services is necessary to improve efficiency in both public and private health care delivery systems.

 Concerns over program inefficiency and variation in health care spending led to a 2013 Institute of Medicine (IOM) report that documented the extent of variation in service utilization and expenditures in Medicare. The IOM report found that at least 36% percent of variation in regional spending was unwarranted—not explained by differences in disease burden or severity among patients.

 The IOM report also raised serious concerns with services provided in postacute and long-term care settings, finding that variation in postacute care spending alone accounted for 73% of the total observed variation in Medicare spending.

Home health care is a critical component of postacute and long-term care services in the United States, which, despite extensive variation, remains understudied.

 As of 2017, more than 12,000 Medicare-certified home health agencies participated in the program, delivering care to more than 3 million beneficiaries.

 The number of beneficiaries is expected to increase due to the aging US population and policy changes by CMS.

Medicare home health expenditures increased 113% from $8.5 billion in 2000 to $18.1 billion in 2016 in part due to the implementation of Home Health Resource Groups (HHRG), a prospective payment system that replaced the Medicare fee-for-service (FFS) mechanism previously used to reimburse home health agencies.

 With the increasing number of participating home health agencies, beneficiaries served, and expenditures under HHRG, information describing the extent of variation in home health care is needed to help policy makers and other stakeholders identify potential reforms.

This study describes regional variation in US home health spending to better understand unwarranted variation. Additionally, the study seeks to identify the sources of variation in home health care spending to inform policy makers on strategies to reduce unwarranted variation.

The conceptual framework is based on the literature describing home health utilization, which is a function of patient, home health agency, and community characteristics. In addition to age and gender, evidence has shown that dual-eligible Medicare and Medicaid beneficiaries use more home health resources.

 The CMS Hierarchical Conditional Category (HCC) risk score, a score assigned to patients based on health status and health conditions, is associated with health care consumption and used to adjust payment for private insurance plans that cover Medicare beneficiaries under Medicare Part C.

 Agency characteristics include ownership type (for profit, not for profit, government) and whether market entry took place during the era of the HHRG prospective payment system implemented in 2000.

 The number of primary care physicians in the community affects coordination between physicians and home health professionals and the timeliness of care received.

 Competition among home health agencies, skilled nursing facilities, and hospitals also affects patient choice about the use of home health vs other long-term care services.

This is a retrospective study design that aggregated home health agency data at the county level to examine variation. Analysis of service variation commonly relies on specific geographic areas, such as hospital referral regions, health service areas, or counties.

 The county was used as the geographic unit for analysis because the majority of Medicare home health beneficiaries receive care from agencies in their home county.

 The study sample consists of all Medicare-certified home health agencies serving 11 or more beneficiaries in 2016 across all 50 states and the District of Columbia.

Data sources included the 2016 Medicare Provider Utilization and Payment Data files: the Public Use File Home Health Agencies (PUF HHA) file, the Provider of Services (POS) file, the Home Health Compare (HHC) file, and the Area Health Resources File (AHRF). The PUF HHA file contains agency-level information, including provider identification number, the total Medicare standard payment amount for beneficiaries who receive at least 5 home health visits during their episode of care (non–Low Utilization Payment Adjustment [non-LUPA] beneficiaries), and summarized characteristics of the beneficiaries per home health agency. These include the mean age of beneficiaries, the percentage of dual-eligible beneficiaries, and the mean HCC risk score for patients served. The HHC file provides the agency’s initial date of the contract with CMS and ownership type. The AHRF provides state and county Federal Information Processing Standards (FIPS) codes for each county and details for county-level community characteristics, such as the number of primary care physicians, nursing home beds, and long-term hospital beds. Medicare wage adjustment per county based on Social Security Administration (SSA) state and county codes are available in the Medicare Wage Adjustment files.

The dependent variable is county-level Medicare standard home health expenditure per home health beneficiary because it eliminates geographic factors incorporated by Medicare to adjust provider payment. Per-beneficiary payments were calculated by aggregating agency-level Medicare standard payment amounts at the county level as the numerator and agency-level unique Medicare non-LUPA home health beneficiaries as the denominator.

Quintiles of county-level spending per Medicare home health beneficiary and how quintile assignment relates to the characteristics of patients, providers, and the community are the key measures of interest. Home health agencies behave differently based on when they entered the market in relation to the implementation of HHRG in 2000, as profitable practices in the prospective payment system differed from those in the previous FFS payment system.

 Thus, a measure of tenure as the percentage of home health agencies entering the market before 2000 in each county is included. The percentages of agencies that were government owned and for profit per county, as well as an indicator of agencies operated as part of home care chains, were drawn from the POS file. A link between the wage adjustment file and the PUF HHA data set was created by utilizing a crosswalk between SSA and FIPS codes. For community characteristics, a county-level Herfindahl-Hirschman Index (HHI) of competition was calculated as the sum of the squared market share based on the number of home health beneficiaries served by each agency. Finally, the number of primary care physicians per 1000 population, the number of nursing home beds and long-term hospital beds per 1000 population, and county-level median household income were included at the county level.

Counties were divided into quintiles based on Medicare home health expenditures per home health beneficiary. Counties in quintile 1 had the lowest expenditures; those in quintile 5 had the highest. Additionally, the coefficient of variation (COV) and the ratio of the 90th to 10th percentile for all variables were used to analyze variation in expenditures within and across each quintile.

Ordinary least square regression models were used to assess factors associated with geographic variation in home health expenditure per home health beneficiary. The first model included only 4 dummy variables for counties in quintiles 2 to 5, with those in quintile 1 serving as the reference group. Each iteration of the model successively added patient, agency, and community characteristics. The changes in 

 in each subsequent model show how much variation in home health expenditure per beneficiary is explained by adding patient, agency, and community characteristics.

Counties in each quintile include state effects that influence expenditures. To estimate geographic variation resulting from state-level fixed effects, we excluded the dummy variable for quintiles and added patient, agency, and community characteristics successively and analyzed models with and without state-level fixed effects. Statistical analysis was conducted with Stata 14.2 (StataCorp).

The PUF HHA file contains information on 10,046 home health agencies that served 11 or more patients in the United States in 2016. A total of 1925 of 3141 counties in the United States had at least 1 agency in the PUF HHA file and were included in our analysis. Counties not included in the study were more likely to be rural, with lower population levels and lower median incomes.

 presents the mean of county-level Medicare home health expenditures per home health beneficiary and county-level patient, agency, and community characteristics of the study sample across quintiles of Medicare home health expenditure. On average, overall Medicare expenditures were $5050 per home health beneficiary, ranging from $3440 in quintile 1 to $7305 in quintile 5. The within-quintile 90th-to-10th-percentile ratio of expenditures was 2.5, and the same ratio for the HCC scores was 1.13. Approximately 30.4% of patients were dual-eligible Medicare and Medicaid beneficiaries, representing 25.8% of beneficiaries in quintile 1 vs 37.6% in quintile 5. Overall, non-White beneficiaries comprised 14.6% of the sample, but this ranged from 8.7% in quintile 1 to 22.5% in quintile 5. Agencies included in the study had an mean HCC score of 2.1. Across quintiles, agency HCC score ranged from a low of 2.0 in quintile 1 to a high of 2.2 in quintile 4. The majority (50.3%) of agencies in the study were for profit; 17.1% were government agencies. Major differences were observed across quintiles in agency ownership. Counties in quintile 1 had the lowest percentage of for-profit agencies (15.6%) but had the highest percentage of government-owned agencies (35.1%). Counties in quintile 5 had the highest percentage of for-profit agencies (81.6%) but the lowest percentage of government-owned agencies (6.7%). Overall, tenured agencies that entered the market before HHRG implementation made up approximately 40% of the study sample; they were 43.3% of agencies in quintile 1 and 34.0% of agencies in quintile 5. Overall, 22.5% of agencies in the sample operated as branches of home health chains, with the highest proportion of chain agencies in quintile 5 (30.5%) and the lowest in quintile 1 (11.2%). Medicare wage adjustments were higher in lower quintiles, with quintile 1 at a wage adjustment of 0.90 and quintile 5 at 0.81. Counties in the study had an mean of 0.2 long-term hospital beds, 0.6 physicians, and 0.4 nursing beds per 1000 population. Median household income was $50,792, and the mean HHI score was 7146.

 provides a visualization of the extent of variation in county-level home health expenditure per beneficiary by quintiles. Intraquintile variation was low, with quintiles 1 (COV = 0.1) and 5 (COV = 0.14) exhibiting the highest variation. However, the overall COV for mean expenditures per beneficiary in all study counties was 0.3, indicating substantial overall variation across quintiles. The overall 90th-to-10th-percentile ratio of 2.5 indicates that the 90th percentile mean expenditure is 2.5 times that of the 10th percentile. 

 adds context to the results, identifying counties by quintile of per-patient expenditure by color on a map. The South, particularly Texas, Oklahoma, and Louisiana, has a higher concentration of high-expenditure counties. The New England and West Coast regions have higher concentrations of low-expenditure counties. Counties colored in gray were not included in the study due to a lack of eligible agencies for analysis.

 provides results from models adjusted for selected variables based on our conceptual framework. Coefficients are first presented in dollar spending for each quintile of per-patient expenditure in an unadjusted model and are then adjusted as we add patient, agency, and community characteristics (the results are available in 

]). In the unadjusted model, unexplained spending differences ranged from $758.28 (95% CI, $686.83-$829.73) between quintiles 1 and 2 up to $3864.70 (95% CI, $3793.25-$3936.16) between quintiles 1 and 5, with an 

 remained unchanged after adding beneficiary, agency, and community characteristics to the model.

State policies and other characteristics likely influence expenditures for home health at the county level. To estimate state effects on variation, we excluded dummy variables for county quintiles and applied the models with and without state fixed effects (

 between models with (0.50) and without (0.16) state fixed effects changed to 0.58 and 0.39 once agency and community characteristics were added, representing a percent difference reduction from 212.5% to 48.7%. However, the full model with state fixed effects explained only 58% of the variation (the results are available in 

The IOM report on spending variation identified postacute care as the primary driver of spending inefficiency in Medicare, with postacute care accounting for approximately 70% of the variation in patient-level Medicare spending. The home health industry is an integral component of postacute care for Medicare beneficiaries and provides services to 3.5 million beneficiaries annually through more than 12,000 contracted home health agencies. This study provides new information on the extent of spending variation that exists among home health beneficiaries. A prior study suggested that the source of variation in home health services utilization stems from differences in organizational behavior, local resources, or Medicaid factors,

 but in the present study, these measures had a marginal impact on explaining variation. Excluding the 3 states with the highest rates of variation (Texas, Oklahoma, and Louisiana) resulted in a drop in the 90th-to-10th-percentile ratio from 2.5 to 2.0, indicating persistent unexplained variation. And although these 3 high-utilization states opted not to expand Medicaid, a sensitivity analysis exploring the possible impact of Medicaid expansion on home health utilization showed that Medicaid expansion status was not statistically significant in state-level fixed effects or random effects models. After adjusting for patient, agency, and community factors, a difference of more than $2500 remained between per-beneficiary home health expenditures in quintiles 1 and 5, and more than 40% of the variation remained unexplained by the models in the study, an indication of waste and inefficiency in the home health care delivery system.

Several characteristics of beneficiaries, physicians, and agencies likely contribute to this observed variation. To receive services, the Medicare Home Health Benefit requires beneficiaries to meet 3 criteria: being homebound, requiring intermitted skilled care, and receiving a physician referral through a face-to-face encounter assessment.

 Beneficiaries bear no cost sharing and can receive unlimited 60-day episodes of home health care with physician recertification.

 Without beneficiaries sharing financial responsibility for episodes of home health care, cost does not influence beneficiary decision-making about whether another episode of care is needed, what actions they can take themselves to improve their conditions, and what home health providers can do for them.

Although physicians are required to conduct a face-to-face assessment in order to refer their patients for home health care,

 recent evidence shows that the majority of physicians spend less than 1 to 2 minutes completing the referral form, do not change the referral form once home health professionals submit the renewal certification, and fail to ask home health professionals to clarify any information in the form.

 This physician certification mechanism leaves room for home health agencies to induce unnecessary demand.

At the agency level, home health agencies require only small capital assets to enter and operate in the market and can easily adjust operating systems to maximize profit margins.

 Cabin and colleagues found that for-profit agencies were more costly but provided lower quality of care compared with not-for-profit agencies.

 Under the HHRG, Kim and Norton also found that for-profit agencies that entered the market after 2000 were more financially incentivized to provide therapy visits that yielded high margins than agencies established before the HHRG implementation.

 The success of these new market entrants influenced for-profit peers to adopt similar practice patterns and pursue profitable therapy visits. In addition to for-profit agencies and peer effects, medical fraud is an issue in the home health industry. According to a report by the US Government Accountability Office, home health agencies exhibited the highest rate of medical fraud among all types of health care providers, accounting for more than 40% of medical fraud in the nation in 2010. Although fraud may contribute to unwarranted variation,

 it is not a major source of variation in health care delivery identified by the IOM.

There are limitations to the study. First, the study analyses rely on data from just 1925 of 3141 US counties due to limitations in the PUF HHA file, which suppresses agencies providing services to 10 or fewer patients in the calendar year. Should the behavior of agencies in counties excluded be different from that of those in our study, our results would not generalize to them. Second, our data include only patient risk factors for gender, race, age, dual eligibility, and CMS HCC score. Other health and social risk factors, such as measures of activities of daily living and the capability and availability of informal caregivers, may affect how often home health professionals visit their patients,

 for which our models could not control. Finally, the county, based on the location of the home health agencies, is our geographic unit of measurement. Although the majority of beneficiaries seek care from home health agencies located in their residential county,15 our data do not allow us to distinguish expenditures at the patient level, and some beneficiary spending will be captured in the county of the home health agency rather than their county of residence.

Despite these limitations, our findings have policy implications. First, the Medicare program is a primary payer of home health for older patients and has the purchasing power to set payment rates. However, states regulate the home health market, which influences agency practice patterns. For example, states with certificate-of-need regulations consume less home health care and have lower growth in home health expenditures than those without.

 Findings from this study indicate a large effect of state regulatory policies and characteristics on overall spending, with approximately 20% of the observed variation attributable to state fixed effects. These findings indicate that the Medicare program should work with states to address geographic variation through market regulation.

Second, although face-to-face physician assessment encounters are required by policy, evidence in the literature indicates that the physician referral system should be strengthened, either through incentivizing physicians to perform more meaningful assessments or through assignment of legal responsibility to physicians to certify the referral process. Finally, for individual beneficiaries, increasing cost-consciousness through co-pays when accessing the Medicare Home Health Benefit—as recommended by the Medicare Payment Advisory Commission to CMS

—could reduce unnecessary preference-sensitive home health care utilization.

The demand for home health care is expected to continue to grow, given both the preference to stay home and the changing demographics of the country.

 Reducing unwarranted variation is key to strengthening the Medicare home health care benefit. In 2020, CMS implemented a new payment system, the Patient Driven Groupings Model, which eliminates the number of therapy visits from the payment equation.

 We recommend strengthening the physician referral system, adding co-pays for each episode of home health care, and improving collaboration between states and the Medicare program to ensure that the home health care delivery system provides sustainable, efficient, high-quality care to beneficiaries in need.

 Department of Health Policy and Management, College of Public Health, University of Arkansas for Medical Sciences (RFS, ALML, HFC, JMT), Little Rock, AR.

 The authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

 Concept and design (RFS, HFC, JMT); acquisition of data (RFS); analysis and interpretation of data (RFS, HFC, JMT); drafting of the manuscript (RFS, ALML, HFC, JMT); critical revision of the manuscript for important intellectual content (RFS, ALML, HFC, JMT); statistical analysis (RFS, HFC); administrative, technical, or logistic support (ALML, JMT); and supervision (HFC, JMT).

 Robert F. Schuldt, PhD, University of Arkansas for Medical Sciences, 4301 W Markham St, Little Rock, AR 72205. Email: Rfschuldt@uams.edu.

Variation in Health Care Spending: Target Decision Making, Not Geography

. The National Academies Press; 2013. doi:10.17226/18393

2. Fisher ES, Wennberg DE, Stukel TA, Gottlieb DJ, Lucas FL, Pinder EL. The implications of regional variations in Medicare spending. part 1: the content, quality, and accessibility of care. 

. 2003;138(4):273-287. doi:10.7326/0003-4819-138-4-200302180-00006

3. Fisher ES, Wennberg DE, Stukel TA, Gottlieb DJ, Lucas FL, Pinder EL. The implications of regional variations in Medicare spending. part 2: health outcomes and satisfaction with care. 

. 2003;138(4):288-298. doi:10.7326/0003-4819-138-4-200302180-00007

Tracking Medicine: A Researcher’s Quest to Understand Health Care. 

5. Talaga SR. Medicare home health benefit primer: benefit basics and issues. Federation of American Scientists. March 14, 2013. Accessed January 10, 2020. https://fas.org/sgp/crs/misc/R42998.pdf

6. Newquist DD, DeLiema M, Wilber KH. Beware of data gaps in home care research: the streetlight effect and its implications for policy making on long-term services and supports. 

. 2015;72(5):622-640. doi:10.1177/1077558715588437

7. Medicare Payment Advisory Commission. Home health care services. In: 

Report to the Congress: Medicare Payment Policy

. Medicare Payment Advisory Commission; 2019:225-248. Accessed December 4, 2019.
https://www.medpac.gov/wp-content/uploads/import_data/scrape_files/docs/default-source/reports/mar19_medpac_ch9_sec_rev.pdf

8. Knickman JR, Snell EK. The 2030 problem: caring for aging baby boomers. 

. 2002;37(4):849-884. doi:10.1034/j.1600-0560.2002.56.x

9. Joynt Maddox KE, Chen LM, Zuckerman R, Epstein AM. Association between race, neighborhood, and Medicaid enrollment and outcomes in Medicare home health care. 

. 2018;66(2):239-246. doi:10.1111/jgs.15082

10. Pope GC, Ellis RP, Ash AS, et al. Diagnostic cost group hierarchical condition category models for Medicare risk adjustment. CMS. December 21, 2000. Accessed March 13, 2020. https://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-Reports/Reports/downloads/pope_2000_2.pdf

11. Cabin W, Himmelstein DU, Siman ML, Woolhandler S. For-profit Medicare home health agencies’ costs appear higher and quality appears lower compared to nonprofit agencies. 

. 2014;33(8):1460-1465. doi:10.1377/hlthaff.2014.0307

12. Kim H, Norton EC. Practice patterns among entrants and incumbents in the home health market after the prospective payment system was implemented. 

. 2015;24(suppl 1):118-131. doi:10.1002/hec.3147

13. Boyd CM, Leff B, Bellantoni J, et al. Interactions between physicians and skilled home health care agencies in the certification of Medicare beneficiaries’ plans of care: results of a nationally representative survey. 

. 2018;168(10):695-701. doi:10.7326/M17-2219

14. Li Q, Rahman M, Gozalo P, Keohane LM, Gold MR, Trivedi AN. Regional variations: the use of hospitals, home health, and skilled nursing in traditional Medicare and Medicare Advantage. 

. 2018;37(8):1274-1281. doi:10.1377/hlthaff.2018.0147

15. Franco SJ. Medicare home health care in rural America. 

16. Welch HG, Wennberg DE, Welch WP. The use of Medicare home health care services. 

. 1996;335(5):324-329. doi:10.1056/NEJM199608013350506

17. Medicare & home health care. CMS. Updated September 2020. Accessed November 9, 2021. https://www.medicare.gov/Pubs/pdf/10969-Medicare-and-Home-Health-Care.pdf

18. Patient Protection and Affordable Care Act, Pub L No. 111-148 (2010) Sec. 6407. Accessed March 13, 2020. https://www.congress.gov/111/plaws/publ148/PLAW-111publ148.pdf

19. Medicare Payment Advisory Commission. Home health services. In: 

. Medicare Payment Advisory Commission; 2011:173-199. Accessed January 10, 2020. https://www.medpac.gov/wp-content/uploads/import_data/scrape_files/docs/default-source/reports/Mar11_Ch08.pdf

20. Health care fraud: types of providers involved in Medicare, Medicaid, and the Children’s Health Insurance Program cases. US Government Accountability Office. September 2012. Accessed March 13, 2020. https://www.gao.gov/assets/650/647849.pdf

21. Osakwe ZT, Larson E, Andrews H, Shang J. Activities of daily living of home healthcare patients. 

. 2019;37(3):165-173. doi:10.1097/NHH.0000000000000736

22. Cho E, Kim EY, Lee NJ. Effects of informal caregivers on function of older adults in home health care. 

. 2013;35(1):57-75. doi:10.1177/0193945911402847

23. Polsky D, David G, Yang J, Kinosian B, Werner R. The effect of entry regulation in the health care sector: the case of home health. 

. 2014;110:1-14. doi:10.1016/j.jpubeco.2013.11.003

24. Rahman M, Galarraga O, Zinn JS, Grabowski DC, Mor V. The impact of certificate-of-need laws on nursing home and home health care expenditures. 

. 2016;73(1):85-105. doi:10.1177/1077558715597161

25. Medicare Payment Advisory Commission. Home health care services. In: 

. Medicare Payment Advisory Commission; 2017:229-253. Accessed September 10, 2019. https://www.medpac.gov/wp-content/uploads/import_data/scrape_files/docs/default-source/reports/mar17_medpac_ch9.pdf

26. Fixing to stay: a national survey on housing and home modification issues. AARP. May 2000. Accessed March 22, 2019. https://assets.aarp.org/rgcenter/il/home_mod.pdf

Articles

This study attempts to identify the sources of the significant 2.5-fold variation found in home health expenditures, a possible indicator of inefficiency and waste.

Geographic Variation in Medicare Home Health Expenditures

We appreciate the opportunity to address the RTI team’s main concerns of our study. Our paper demonstrated that CMS Hierarchical Condition Categories (HCC) risk scores are related to patient experience in home health.

 These findings suggest that HCC risk scores should be considered when evaluating the performance of home health agencies. However, the RTI team raised 2 concerns regarding the validity of agency-level HCC risk scores. First, HCC risk scores are prone to upcoding by providers across the spectrum, resulting in an upward bias. Statistically, upcoding will change the HCC risk score but not the SD. Our findings are based on the SD of HCC risk scores and therefore unlikely to be affected by upcoding. Secondly, the RTI team indicated that agency-level HCC risk scores likely include outliers, whereas the random sampling used in the Home Health Care Consumer Assessment of Healthcare Providers and Services (HHCAHPS) is unlikely to be affected by outliers, due to the exclusion of patients with certain conditions.

 The exclusion criteria are not based on the HCC risk scores. Additionally, most patients treated by agencies still remain in the sampling pool, regardless of their outlier scores. Theoretically, a randomizing process that selects patients from each agency will yield the characteristics of the sample that are similar to those of the patients cared by the agency. Thus, the outliers at the agency level of HCC risk scores and the outliers of HCC risk scores from the randomized HHCAHPS sampling should not differ significantly.

The RTI team also identified 2 methodological issues: (1) the reporting of excessive correlation as it relates to the authors’ combined use of state fixed effects and race variables and (2) our use of controls for agency profit/nonprofit status and term of certification by Medicare and how these factors are reflected in the agency’s public reporting. The correlation of most of our independent variables was less than 0.20, with the exception of ownership and tenure years with CMS. Therefore, the correlation in our study is not a concern (

). States vary in racial composition, but the correlation between racial/ethnic composition and state fixed effects is not our primary focus. States certify and regulate home health agencies and through these policies affect the practice of home health agencies.

 Additionally, agencies with different ownership and CMS tenure years respond to public reporting differently.

 Thus, controlling for those confounders (ie, state fixed effects, ownership of agency, and number of tenure years) is necessary. We reanalyzed the data by excluding ownership, number of years certified by CMS, and state fixed effects and then estimated the association between HCC risk scores and patient experience (

). The coefficients in the model without those variables remain significant and are larger than those in our original paper.

For the reporting issue about the effect size, the RTI team recommended 3 levels using a small size (difference of 1 point), a medium size (3 points), and a large size (≥5 points). Our presentation is based on the association between the change of 1 SD in HCC scores and the change in the percent patient experience. We believe that we should provide the association instead of arbitrarily deciding a small, medium, or large effect size for the readers.

Finally, although the RTI team is unable to adjust HCC scores for patient experience due to limitations of the data and time constraints, our findings, at least, provide evidence for stakeholders that HCC risk scores can influence patient experience at the agency level.

 Department of Health Policy and Management, College of Public Health, University of Arkansas for Medical Sciences (HFC, JMT, RFS), Little Rock, AR; Division of Public Health Sciences, Department of Surgery, Washington University School of Medicine (FW), St. Louis, MO.

 Concept and design (HFC, FW); acquisition of data (HFC, RFS); analysis and interpretation of data (HFC, JMT, FW); drafting of the manuscript (HFC, JMT); critical revision of the manuscript for important intellectual content (HFC, JMT, RFS, FW); statistical analysis (HFC, RFS); and provision of patients or study materials (HFC, RFS).

 Hsueh-Fen Chen, PhD, University of Arkansas for Medical Sciences, 4301 W Markham St, Mail Slot 820-12, Little Rock, AR 72205. Email: hchen@uams.edu.

1. Chen HF, Tilford JM, Wan F, Schuldt R. CMS HCC risk scores and home health patient experience measures. 

2. Home Health Care CAHPS survey: protocols and guidelines manual. Home Health CAHPS website. homehealthcahps.org/Portals/0/PandGManual.pdf. Published January 2019. Accessed December 16, 2019.

3. Kim H, Norton EC. Practice patterns among entrants and incumbents in the home health market after the prospective payment system was implemented. 

. 2015;24(suppl 1):118-131. doi: 10.1002/hec.3147.

4. Polsky D, David G, Yang J, Kinosian B, Werner R. The effect of entry regulation in the health care sector: the case of home health. 

. 2014;110:1-14. doi: 10.1016/j.jpubeco.2013.11.003.

5. Jung K, Shea D, Warner C. Agency characteristics and changes in home health quality after Home Health Compare. 

. 2010;22(4):454-476. doi: 10.1177/0898264310362540.

The authors of “CMS HCC Risk Scores and Home Health Patient Experience Measures” respond to a letter to the editor.

Reply to “Risk Adjustment in Home Health Care CAHPS”

 To understand the association between agency-level CMS Hierarchical Condition Categories (HCC) risk scores and patient experience measures for home health.

 We extracted variables from the 2014 Medicare Provider Utilization and Payment Data for Home Health Agencies and Home Health Compare file. We applied fixed-effects models for the analyses. Our dependent variables included both global and composite patient experience measures. The 2 global patient experience measures were the patient’s overall rating of care provided by the agency (

) and the patient’s willingness to recommend the home health agency to others (

). The 3 composite patient experience measures were how often the patient felt the provider gave care in a professional way (

), how well the home health team communicated with the patient (

), and whether the home health team discussed medicines, pain, and home safety with the patient (

 Increased agency-level CMS HCC risk scores were negatively associated with all patient experience measures: 

 ≤.001). Several covariates, including the percentage of racial/ethnic minority beneficiaries, ownership of the agency, and number of tenured years with the Medicare program, were significantly associated with patient experience measures.

 A negative association exists between CMS HCC risk scores and patient experience measures. To avoid unintended consequences, patient experience measures need further risk adjustment under the CMS 5-star patient survey rating system and the Home Health Value-Based Purchasing pilot program.

The Home Health Compare website uses patient experience measures to construct a 5-star patient survey rating system for the public. These measures are also quality performance metrics used in the Home Health Value-Based Purchasing (HHVBP) pilot program.

Increases in CMS Hierarchical Condition Categories risk scores were negatively associated with patient experience measures for home health at the agency level.

Evaluating the practice patterns of home health agencies and monitoring access to care for racial/ethnic minorities and beneficiaries with complicated clinical conditions under the 5-star patient survey rating system and HHVBP pilot program is recommended.

Patient experience is a critical quality dimension of healthcare and is also associated with patient outcomes, such as patient safety and readmissions.

 In 2010, CMS used the Home Health Care Consumer Assessment of Healthcare Providers and Systems (HHCAHPS) survey to assess patient experience for home health care.

 In 2016, CMS used the results from the HHCAHPS survey to create a 5-star patient survey rating system and made it publicly available.

 Patient experience is also part of quality metrics in the Home Health Value-Based Purchasing (HHVBP) pilot program that financially rewards or penalizes home health agencies in 9 states.

In the United States, about 12,000 home health agencies deliver care to approximately 3.5 million Medicare beneficiaries at their residences.

 Medicare home health beneficiaries are homebound and often experience polypharmacy.

 They have, on average, 4.2 diagnostic conditions and have a higher prevalence of limitations in activities of daily living and cognitive function than general Medicare beneficiaries.

 Most of them receive assistance from informal caregivers to meet their medical and daily needs.

 Home health providers not only conduct professional care (eg, physical therapy and wound care) but also educate beneficiaries and their informal caregivers about how to conduct safe care (eg, proper use of medications and oxygen inhalation) and coordinate necessary resources to meet patients’ needs and expectations at home.

Evidence from surveys for health plans, hospitals, community health centers, and primary care physicians shows that several factors affecting patient experience are beyond health providers’ control. The most

common factors are patient sociodemographics, self-reported health status, depression, survey language, mode of survey (eg, mail or telephone interview), and whether the survey was completely by a proxy.

 Other studies focusing on hospital care found that patients with high severity of illness and complicated conditions were likely to have lower ratings of patient experience than their counterparts. They concluded that risk factors used in patient experience for hospital care insufficiently adjusted for clinical conditions that affected patient experience ratings.

In home health, CMS and the HHCAHPS coordination team reported that the factors associated with patient experience are (1) patient sociodemographics, including age and education; (2) self-reported health status; (3) whether patients live alone; (4) self-reported mental/emotional status; (5) non-English survey response; (6) a diagnosis of schizophrenia or dementia/cerebral degeneration; and (7) whether the survey was completely by a proxy.

 These 7 factors are currently used to adjust patient experience for the 5-star patient survey rating system and the HHVBP pilot program. A study also found that minorities had lower ratings of patient experience for their home health care.

Medicare home health beneficiaries are among the most vulnerable Medicare populations. Whether the risk factors currently used by CMS adequately adjust patient experience for home health care is largely unknown. Our study aimed to fill this critical knowledge gap by examining the interplay between agency-level CMS Hierarchical Condition Categories (HCC) risk scores (details in Methods section) and the patient experience metrics. We used the agency-level CMS HCC risk score as a proxy variable for the clinical and functional conditions of home health beneficiaries within home health agencies. Associations between patient experience measures and agency-level CMS HCC risk scores from our study provide critical information about whether the current risk-adjusted model for patient experience measures needs modification.

We merged 2 national Medicare databases at the home health agency level, the 2014 Medicare Provider Utilization and Payment Data for Home Health Agencies (PUPDHHA) and the 2014 Home Health Compare file.

 The PUPDHHA provides several agency-level variables based on 100% of Medicare enrollment and fee-for-service claims data during the 2014 calendar year.

 These variables include an agency-level HCC risk score based on the HCC risk scores from all of the beneficiaries cared for by a particular home health agency, provider identification number, number of distinct beneficiaries without a low-utilization payment episode (with ≤4 home health visits during an episode of home health care, namely, up to 60 days), and number of beneficiaries from different racial/ethnic groups.

The Home Health Compare file is updated quarterly by CMS.

 It consists of provider identification numbers and patient experience measures extracted from the HHCAHPS survey. The Home Health Compare file also provides characteristics of home health agencies, including ownership and CMS certification year. We extracted agency-level patient experience measures based on data collected from January 1, 2014, through December 31, 2014.

We used a cross-sectional study design in which the home health agency was the unit of analysis. The study sample consisted of Medicare-certified home health agencies in 50 states and the District of Columbia. Home health agencies with fewer than 60 patients were excluded because they were exempted from the HHCAHPS survey.

The dependent variables are the 5 risk-adjusted agency-level patient experience measures from the 2 global questions and 3 composite measures extracted from CMS Home Health Compare. The HHCAHPS survey has 2 global questions; they include the “overall rating of care” provided by the home health agency (hereafter termed 

) and “patient willingness to recommend the home health agency to family or friends” (hereafter termed 

 in the survey ranges from 0 to 10. CMS reports the percentage of 

for an agency based on the percentage of patients who gave their home health care a rating of 9 or 10. The choices in the survey for 

 include (1) definitely no, (2) probably no, (3) probably yes, and (4) definitely yes. The percentage of 

 at the agency level is the number of patients who answered “definitely yes” divided by the total number of patients.

For the composite measures, CMS uses the data from the other 17 patient experience questions from the HHCAHPS survey. Each of these 3 measures is calculated from 4 or more topically related survey questions.

 The resulting composite measures include the following: “how often the home health team gives care in a professional way” (hereafter termed 

), “how well the home health team communicates with patients” (hereafter termed 

), and whether or not the “home health team discuss[ed] medicines, pain, and home safety with patients” (hereafter termed 

). The HHCAHPS survey website provides detailed information for the composite measures.

The risk-adjusted global and composite measures for patient experience at the agency level are measured as a percentage. A higher percentage of patient experience measures indicates that the patients perceive higher-quality care from home health agencies.

The key independent variable is the agency-level HCC risk score, extracted from PUPDHHA.

 The agency-level HCC risk score is the sum of CMS HCC risk scores from individual Medicare home health beneficiaries divided by the total number of Medicare home health beneficiaries for an individual home health agency.

 CMS constructs a CMS HCC risk score for an individual Medicare home health beneficiary as a risk factor to calculate a beneficiary’s likelihood of unplanned hospitalization and emergency department visits within 60 days of home health care for public reporting.

 The CMS HCC risk score for an individual Medicare home health beneficiary is a function of the beneficiary’s age, gender, original reason for Medicare entitlement, prior care setting, clinical conditions from CMS HCC, and their interaction terms. The CMS website provides detailed information about the CMS HCC risk score for home health beneficiaries.

In addition to the average CMS HCC risk score at the agency level, previous studies show that beneficiaries of different races/ethnicities rate their respective patient experience surveys differently.

 Therefore, we included variables describing the percentage of beneficiaries who were African American, Hispanic, and of other racial group, which includes beneficiaries who were not white, African American, or Hispanic. Finally, because the characteristics of home health agencies are associated with the quality of care provided,

 we included 2 dummy variables for ownership of a home health agency (ie, not-for-profit and public home health agencies, with for-profit home health agencies as the reference group) and the number of years that home health agencies had been certified by the Medicare program in the analytical models.

Home health agencies practice within a given state, facing that state’s regulations, which may differ from those of other states. Additionally, there is high variation in home health utilization across states, with $176 per beneficiary per year for the states at the 10th percentile and $866 per beneficiary per year for those at the 90th percentile.

 We applied fixed-effects models at the state level, with robust standard errors for each patient experience measure to account for differences in regulations and other state characteristics that affect the practice of home health agencies. Stata 14.2 (StataCorp; College Station, Texas) was used for data management and analyses. We used xtreg for the fixed-effects model.

There were 7756 home health agencies with data for patient experience measures available in the Home Health Compare file. Due to missing data, such as the agency-level CMS HCC risk score or race/ethnicity, the total number of observations included in our analytical models was 7637.

 presents descriptive statistics. For the overall patient experience measures, the average of 

with score 9 or 10 was about 84%, and the average of 

 with “definitely yes” was about 79%. For the composite measures, the averages of 

Risk adjustment for patient experience measures needs to be modified by including the CMS Hierarchical Condition Categories (HCC) risk scores of home health beneficiaries.

J. Mick Tilford, PhD

Articles

Geographic Variation in Medicare Home Health Expenditures

Reply to “Risk Adjustment in Home Health Care CAHPS”

CMS HCC Risk Scores and Home Health Patient Experience Measures

Understanding Factors Associated With Readmission Disparities Among Delta Region, Delta State, and Other Hospitals

Value-Based Insurance Designs in the Treatment of Mental Health Disorders