Medicare’s Hospital Acquired Condition Reduction Program Disproportionately Affects Minority-Serving Hospitals: Variation by Race, Socioeconomic Status, and Disproportionate Share Hospital Payment Receipt

Abstract

Objective:

To assess whether a hospital’s percentage of Black patients associates with variations in FY2017 overall/domain-specific HACRP scores and penalty receipt. Differences in socioeconomic status (SES) and receipt of DSH payments (a marker of safety-net status) were also assessed.

Summary of background data:

In FY2015, Medicare began reducing payments to hospitals with high adverse event rates. Concern has been expressed that HACRP penalties could adversely affect minority-serving hospitals, leading to reductions in resources and exasperation of disparities among hospitals with the greatest need.

Methods:

100% Medicare FFS claims from 2013–2014 identified older adult inpatients, aged ≥65y, presenting for 8 common surgical conditions. Multilevel mixed-effects regression determined differences in FY2017 HACRP scores/penalties among hospitals managing the highest decile of minority patients.

Results:

A total of 695,775 patients from 2,923 hospitals were included. As a hospital’s percentage of Black patients increased, climbing from 0.6% to 32.5% (lowest-vs-highest decile), average HACRP scores also increased, rising from 5.33 to 6.36 (higher values indicate worse scores). Increases in HACRP penalties did not follow the same stepwise increase, instead exhibiting a marked jump within the highest decile of racial minority-serving extent (45.7-vs-36.7%; OR[95%CI]: 1.45[1.42–1.47]). Similar patterns were observed for high DSH (OR[95%CI]: 1.44[1.42–1.47]; absolute difference: +7.4 percentage-points) and low SES-serving (1.38[1.35–1.40]; +7.3 percentage-points) hospitals. Restricted analyses accounting for the influence of teaching status and severity of patient case-mix both accentuated disparities in HACRP penalties when limiting hospitals to those at the highest known penalty-risk (more residents-to-beds, more severe), absolute differences +13.9, +20.5 percentage-points. Restriction to high operative volume, in contrast, reduced the penalty difference, +6.6 percentage-points.

Conclusions:

Minority-serving hospitals are being disproportionately penalized by the HACRP. As the program continues to develop, efforts are needed to identify and protect patients in vulnerable institutions in order to ensure that disparities do not increase.

MiniAbstract

Using 100% Medicare FFS claims, the study demonstrates that hospitals managing a higher proportion of Black patients are being disproportionately penalized by the Hospital Acquired-Condition Reduction Program, HACRP, absolute difference +8.9 percentage-points (OR[95%CI]: 1.45[1.42–1.47]). Similar patterns were observed for high DSH, +7.4 percentage-points (1.44[1.42–1.47]), and low SES-severing, +7.3 percentage-points (1.38[1.35–1.40]), hospitals. Such findings suggest that as the HACRP continues to develop, efforts are needed to identify and protect patients in vulnerable institutions in order to ensure that disparities do not increase.

Introduction

In fiscal-year 2015 (FY2015; October 1, 2014), the Centers for Medicare & Medicaid Services (CMS) began reducing payments to hospitals with high adverse-event rates. Known as the Hospital Acquired Condition Reduction Program (HACRP), the initiative was intended to incentivize hospitals to control preventable adverse events based on hospitalwide pay-for-performance penalties that are in turn based on metrics for patient safety (Domain 1) and nosocomial infections (Domain 2).¹ Since 2014, the program has been estimated to ‘cost’ hospitals and save Medicare >$373 million nationwide per year.¹ Its ability to promote improvements in quality and, thereby, increases in value² is less clear. Early research suggests a direct association between receipt of HACRP penalties and the presence of quality accreditations, an ability to offer more advanced medical services, and being a major teaching hospital.^3–5

Whether HACRP could be disproportionately affecting ‘vulnerable hospitals’ has yet to be assessed. Many researchers have raised concerns that hospitals that routinely provide care to disadvantaged populations, including those from lower socioeconomic status (SES) areas and with a larger percentage of racial minority patients, might be more heavily penalized.^3–5 Prior surgical studies assessing other pay-for-performance programs have shown that payment penalties often land heavily on hospitals that serve a large proportion of vulnerable patients.^6–9 If true of HACRP, this trend has the potential to worsen healthcare disparities by leading to reductions in resources and exasperation of existing differences among hospitals with the greatest need.

The objective of this study was to gauge the impact of Medicare’s HACRP on vulnerable population-serving hospitals by determining the extent to which FY2017 a) overall and domain-specific HACRP quality-metric scores and b) resultant HACRP penalties associate with a hospital’s percentage of racial minority patients (defined as Black versus other race patients based Social Security Administration-defined race in Medicare data in order to account for known limitations of Medicare ethnicity reporting¹⁰). We focused on patients undergoing eight common surgical procedures because most of the performance measures in HACRP apply to surgical patients. As a secondary objective, differences in patient SES and hospital receipt of disproportionate share hospital (DSH) payments (a marker of safety-net status) were also assessed.

Methods

Data source and study population

Records of index inpatient admissions for Medicare patients aged ≥65y were abstracted using 2013–2014 CMS national Medicare Provider Analysis and Review (MedPAR) files. MedPAR contains discharge information for 100% of fee-for-service hospitalizations for Medicare beneficiaries. Included operations consisted of: (1) abdominal aortic aneurism (AAA) repair (International Classification of Diseases, 9^th edition, primary procedure codes: 38.34, 38.44, 38.64, 39.25, 39.71, 39.78), (2) aortic valve repair (AVR; 35.21, 35.22), (3) bariatric surgery (43.82, 43.89, 44.3, 44.31, 44.38, 44.39, 44.68, 44.95, 44.96, 44.99, 44.5), (4) coronary artery bypass graft (CABG) surgery (36.10–36.19), (5) elective colectomy (45.7, 45.71–45.76, 45.79, 45.8, 45.81–45.83, 17.31–17.39), (6) lung resection (32.20, 32.29, 32.3, 32.30, 32.39, 32.4, 32.41, 32.49, 32.5, 32.50, 32.59, 32.6, 32.9), (7) mitral valve repair (MVR; 352.3, 352.4), and (8) rectal resection (48.40–48.49, 48.50–48.59, 48.62, 48.63, 48.65, 48.69). They were chosen to represent a broad array of common surgical procedures performed at most hospitals treating Medicare patients. Identified patient records were matched to risk-adjusted FY2017 HACRP scores and CMS acknowledgement of HACRP penalty receipt available for each hospital from Hospital Compare, hospital-level data contained within the 2013–2014 Medicare Hospital Impact files, and additional hospital-level data obtained from the 2013–2014 American Hospital Association Annual Survey Database. Receipt of DSH payments was derived from information reported in the 2013–2014 Medicare Hospital Impact files. Details of specific HACRP metrics and composite scores are available online.¹

Explanatory and outcome variables

Racial minority-serving hospitals were defined as those managing the highest decile of Black patients.¹¹ Separate values were calculated for surgical patients overall and for each operation type. Differences in HACRP scores were initially assessed based on differences in hospital decile of Black patients; however, when variations in penalty receipt were assessed, a more meaningful binary pattern emerged (Figure 1). For this reason, subsequent analyses were dichotomized based on the highest decile of Black patients versus the ‘rest of the population’ (deciles 1–9) in order to operationally define ‘racial minority-serving’ versus ‘non-racial minority-serving’ hospitals. Similar trends were observed when variations in hospital receipt of DSH payments (‘high DSH hospital’) and patient SES (‘low SES-serving hospital’) were assessed. Sensitivity analyses considered the potential for differences in results when decile variations were alternatively modeled as an ordinal series of nine indicator variables.

Figure 1.

Percentage of Black surgical patients managed within a hospital and associated FY2017 Hospital Acquired Condition (HACRP) penalties and scores, overall and by specific domain (Domain 1: patient safety, Domain 2: nosocomial infections), for (A) all included surgical patients and (B) elective colectomy patients

For both secondary explanatory measures, we evaluated differences in outcomes between hospitals in the highest decile and the rest of the population. Surgical patients overall and for each operation type were categorized according to the total percentage of a hospital’s inpatient admissions that received DSH compensation each year—a method used to denote safety-net status among Medicare institutions.⁵ Information on patient SES was obtained by matching patient-level claims with the median age-adjusted annual income of a patient’s residential zip code for each year as reported by the US Department of Treasury, Internal Revenue Service. SES deciles were calculated overall and for each operation type based on the percentage of patients with an annual income <$25,000.

Outcomes derived from Hospital Compare consisted of FY2017 risk-adjusted overall and domain-specific HACRP scores (October 2016-September 2017) and whether or not a hospital received a HACRP payment penalty that year. Within each explanatory variable group, mean HACRP scores and the total percentage of hospitals receiving HACRP penalties were compared.

Statistical analysis

Differences in average HACRP scores and penalty receipt were compared across deciles of (a) racial minority-serving hospitals, (b) high DSH hospitals, and (c) low SES-serving hospitals. Differences between the highest decile and the rest of the population for each factor (a-c) were assessed using hierarchical (mixed-effects) logistic regression models that accounted for clustering of patients within hospitals, yielding odds ratios (OR) and 95% confidence intervals (95%CI). Random intercepts were allowed for each hospital. Absolute percentage-point differences in penalty receipt were also compared. To account for potential differences in geography not accounted for in Medicare’s construct of risk-adjusted HACRP scores, stratified differences in racial minority-serving outcomes were compared among hospitals restricted to each of the four US Census Bureau regions in the United States: Northeast, West, Midwest, and South.

Finally, in order to further explore the hypothesis that marginal changes among the most vulnerable^3–5 and ‘low-quality’¹² institutions could be driving differences between racial minority-serving and non-racial minority-serving hospitals, restricted analyses that compared differences in HACRP payment penalties between racial minority-serving and non-racial minority-serving hospitals were conducted among the subset of hospitals that fell within the highest quartile of: (a) the number of residents-to-beds (a ratio used by CMS to determine indirect medical education payments and a marker of teaching status), (b) complex patient case-mix (taken from a hospital’s case-mix index, a metric calculated by CMS as the sum of all diagnosis-related group weights at a hospital divided by the total number of Medicare discharges), and (c) operative volume. Absolute differences between groups and relative differences from the unrestricted model were calculated in order to determine how the respective ORs compared.

Statistical analyses were conducted using Stata Statistical Software: Version 14.2. The study was deemed exempt from Institutional Review Board review by the University of Michigan.

Results

Racial minority-serving hospitals: Variation in overall and domain-specific HACRP scores

A total of 695,775 patients from 2,923 hospitals met inclusion criteria: 75,132 (10.8%) AAA repair, 58,042 (8.3%) AVR, 51,256 (7.4%) bariatric surgery, 127,552 (18.3%) CABG surgery, 242,361 (34.8%) elective colectomy, 80,940 (11.6%) lung resection, 16,728 (2.4%) MVR, and 43,764 (6.3%) rectal resection. Compared to non-racial minority-serving hospitals (Table 1), racial minority-serving hospitals were, overall, similar. Surgical patients presenting to racial minority-serving hospitals were marginally more likely to be female (49.2 versus 46.9%) and present with a high burden of pre-existing medical conditions (Elixhauser ≥4: 39.5% versus 37.0%), most notably diabetes without chronic complications (28.4% versus 24.8%), deficiency anemias (22.4 versus 18.6%), and hypertension (76.6% versus 74.1%). At the hospital level, racial minority-serving hospitals were more likely to be urban (95.3% versus 91.9%), large hospitals (≥500 beds: 37.8 versus 33.8%) located in the American South (68.2% versus 35.9%).

Table 1.

Overall variations in patient- and hospital-level characteristics comparing racial minority-serving (highest-decile of Black patients) versus non-minority-serving hospitals

	Non-minority serving hospital		Minority serving hospital
	n=626,198		n=69,578
Patient characteristics
Mean age (SD) in years	75.0	6.9	74.4	6.8
Female	293,687	46.9%	34,197	49.2%
Elixhauser count
0–1	128,997	20.6%	12,934	18.6%
2	135,760	21.7%	14,597	21.0%
3	130,500	20.8%	14,528	20.9%
4	99,879	16.0%	11,425	16.4%
5	64,999	10.4%	7,953	11.4%
6+	66,064	10.6%	8,141	11.7%
Comorbidities
Congestive heart failure	39,889	6.4%	5,295	7.6%
Valvular disease	25,674	4.1%	2,512	3.6%
Pulmonary circulation disease	13,401	2.1%	1,809	2.6%
Peripheral vascular disease	103,197	16.5%	10,945	15.7%
Paralysis	7,953	1.3%	1,273	1.8%
Other neurological disorders	32,437	5.2%	3,820	5.5%
Chronic pulmonary disease	163,876	26.2%	17,819	25.6%
Diabetes without chronic complications	155,234	24.8%	19,760	28.4%
Diabetes with chronic complications	25,674	4.1%	3,096	4.5%
Hypothyroidism	95,245	15.2%	8,669	12.5%
Renal failure	88,356	14.1%	10,750	15.5%
Liver disease	10,082	1.6%	1,183	1.7%
Peptic ulcer disease	2,505	0.4%	278	0.4%
AIDS	1,252	0.2%	417	0.6%
Lymphoma	5,260	0.8%	501	0.7%
Metastatic cancer	43,270	6.9%	5,323	7.7%
Solid tumor without metastasis	20,665	3.3%	2,491	3.6%
Rheumatoid arthritis	20,477	3.3%	1,899	2.7%
Coagulopathy	75,394	12.0%	8,231	11.8%
Obesity	88,169	14.1%	9,678	13.9%
Weight loss	57,673	9.2%	7,514	10.8%
Fluid and electrolyte disorders	189,237	30.2%	22,675	32.6%
Chronic blood loss anemia	13,839	2.2%	1,823	2.6%
Deficiency anemias	116,347	18.6%	15,613	22.4%
Psychoses	13,338	2.1%	1,531	2.2%
Depression	54,354	8.7%	4,884	7.0%
Hypertension	463,824	74.1%	53,275	76.6%
Hospital characteristics
ACS-certified trauma hospital	384,611	61.4%	37,001	53.2%
NCI cancer hospital	453,054	72.4%	51,404	73.9%
Urban hospital	575,413	91.9%	66,307	95.3%
Mean full-time residents:bed (SD)	0.16	0.31	0.25	0.38
Mean full-time RN:patient (SD)	1.73	0.62	1.63	0.60
Mean full-time RN:bed (SD)	2.01	0.67	1.89	0.60
Profit status
For profit	85,225	13.6%	9,191	13.2%
Not for profit	492,442	78.6%	51,724	74.3%
Other	48,530	7.8%	8,662	12.5%
Total beds
<200	113,843	18.2%	8,725	12.5%
200–349	169,950	27.1%	19,002	27.3%
350–499	131,001	20.9%	15,585	22.4%
≥500	211,467	33.8%	26,272	37.8%
Region
Midwest	154,170	24.6%	10,917	15.7%
Northeast	127,055	20.3%	9,045	13.0%
South	224,680	35.9%	47,417	68.2%
West	120,230	19.2%	2,192	3.2%

Abbreviations:SD – standard deviation, AIDS – acquired immunodeficiency syndrome, ACS – American College of Surgeons, NCI – National Cancer Institute, RN – registered nurse

As a hospital’s percentage of Black patients increased, overall climbing from 0.6% to 32.5% in the lowest versus the highest decile (Figure 1), average HACRP scores also increased, rising from 5.33 to 6.36 (higher values indicate worse scores). Patient safety, Domain 1, scores started lower and increased more dramatically, jumping from 5.20 to 6.33, while trends for nosocomial infections, Domain 2, mirrored the overall scores, increasing from 5.34 to 6.43. When operation specific trends were assessed, similar patterns emerged. Results for elective colectomy are presented in Figure 1. We opted to highlight the results for elective colectomy given the operation’s large contribution to the overall population size comprising 34.8% of the combined study sample, recognized risk for infectious complications, and historical focus as a target of surgical quality improvement initiatives.¹³

Racial minority-serving hospitals: Variation in HACRP payment penalties

In FY2017, Medicare payment penalties were instituted for hospitals with overall HACRP scores >6.57.¹ This resulted in 28.0–40.5% of patients at hospitals in deciles 1–9 being managed at hospitals that received payment penalties (30.3–40.4% for elective colectomy). Increases in the percentage of patients at penalized institutions did not follow a stepwise increase (Figure 1). Instead, they exhibited a marked jump within the highest decile of racial minority-serving extent: 45.7% (overall) and 51.5% (elective colectomy). Similar patterns were observed for high DSH and low SES-serving hospitals.

Dichotomized differences for each of the three explanatory variables are presented in Figure 2 and Table 2 (operation-specific results: Supplemental Table 1). For surgical patients overall, receipt of HACRP penalties was significantly higher in racial minority-serving hospitals (45.7 versus 36.7%; OR[95%CI]: 1.45[1.42–1.47])—an absolute difference of +8.9 percentage-points. Differences among operation-specific comparisons closely followed suit, demonstrating significantly higher prevalences of penalties in six of the eight operations included. The most pronounced differences were found among MVR (OR[95%CI]: 1.95[1.67–2.28]), bariatric surgery (2.36[2.21–2.51]), and elective colectomy (2.09[2.03–2.15]). No differences were found for AVR and AAA repair cases. Similar results were observed in ordinal sensitivity analyses.

Figure 2.

Differences in HACRP payment penalties comparing the highest decile versus the rest of the population (deciles 1–9): (A) racial minority-serving, (B) high DSH, and (C) low SES-serving hospitals

Table 2.

Overall and restricted differences between the highest decile and the rest of the population

	Highest decile	Deciles 1–9	Absolute difference	_	OR	95%CI
Minority-serving hospitals
Black patients (average %)	32.5%	6.6%	25.9%
HAC penalty (%)	45.7%	36.7%	8.9%		1.45*	1.42	1.47
High DSH hospitals
DSH payments (average %)	53.4%	24.1%	29.3%
HAC penalty (%)	44.1%	36.7%	7.4%		1.44*	1.42	1.47
Low SES-serving hospitals
(median income residential zip)
Patients <$25,000 (average %)	99.4%	8.2%	91.3%
HAC penalty (%)	39.3%	32.0%	7.3%		1.38*	1.35	1.40
Minority-serving hospitals
	Highest decile	Deciles 1–9	Absolute difference	Relative difference from the unrestricted model	OR	95%CI
All minority-serving hospitals	45.7%	36.7%	8.9%	Reference	1.45*	1.42	1.47
Restricted to: Complex-case mix	68.3%	47.9%	20.5%	129.3%	2.35*	2.27	2.43
Restricted to: Major teaching	66.4%	52.5%	13.9%	56.2%	1.79*	1.74	1.84
Restricted to: High volume	58.0%	51.5%	6.6%	−26.5%	1.30*	1.25	1.36
Restricted to: Northeast	58.6%	49.2%	9.4%	5.2%	1.46*	1.39	1.53
Restricted to: West	42.1%	34.2%	7.9%	−11.5%	1.40*	1.27	1.53
Restricted to: Midwest	37.5%	28.7%	8.8%	−1.7%	1.49*	1.42	1.55
Restricted to: South	46.4%	30.7%	15.8%	76.8%	1.96*	1.92	2.00

Odds ratios (OR) and 95% confidence intervals (95%CI) were taken from hierarchical (mixed effects) logistic regression models with a random intercept for each hospital.

High DSH and low SES-serving hospitals: Variation in HACRP payment penalties

Akin to racial minority-serving hospitals, high DSH hospitals were more likely to be penalized than lower DSH hospitals (44.1 versus 36.7%, absolute difference +7.4 percentage-points), as where low SES-serving hospitals (39.3 versus 32.0%, absolute difference +7.3 percentage-points). Odds among hospitals receiving a larger share of DSH payments were 1.44 times higher than those for the rest of the population (95%CI: 1.42–1.47). Odds among lower SES-serving hospitals were 1.38 times as high (95%CI: 1.35–1.40). For high DSH hospitals, stratified differences were significant for the majority of operations. Notable exceptions included a lack of difference in bariatric surgery and significantly better outcomes for high DSH hospitals performing MVR and AVR cases. For low SES-serving hospitals, differences were significant in seven of the eight operation types. Better outcomes were isolated to AAA repair cases.

Restricted analyses

Restricted analyses accounting for the influence of teaching status and severity of patient case-mix both accentuated differences between racial minority-serving and non-racial minority-serving hospitals. Relative to a baseline absolute difference of +8.9 percentage-points between racial minority-serving versus non-racial minority-serving hospitals overall (Table 2), the absolute difference in payment penalties became +20.5 percentage-points when restricted to hospitals with the most complicated case-mix (a relative increase from baseline of 129.3%; restricted group OR[95%CI]: 2.35[2.27–2.43]). For the highest quartile of residents-to-beds, the absolute difference increased to +13.9 percentage-points (a relative increase from baseline of 56.2%; restricted group OR[95%CI]: 1.79[1.74–1.84]). Stratified differences by operation type are reported in Figure 3. Consistent with expectations for managing the most penalty-prone groups of patients,^3–5 hospitals within the restricted analyses experienced increases in baseline penalty rates whether managing a higher percentage of Black patients or not (e.g. overall: 45.7 versus 36.7%; large teaching hospitals: 66.4 versus 52.5%).

Figure 3.

Restricted analyses: Differences in HACRP payment penalties comparing the highest decile of racial minority-serving hospitals versus the rest of the population (deciles 1–9) among hospitals which are: (A) caring for the most complex patients (highest quartile of case-mix index), (B) major teaching institutions (highest quartile of residents-to-beds), and (C) handling the largest operative volume (highest quartile of operative volume)

Restricting analyses to the highest quartile of operative volume, in contrast, mitigated the absolute difference in payment penalties between racial minority-serving and non-racial minority-serving hospitals (a relative drop from baseline of 26.5%; Table 2). The disparity, however, remained significant: absolute difference +6.6 percentage-points, OR(95%CI) of 1.30(1.25–1.36). Restricting hospitals to those with the largest volume did result in increases in the percentage of hospitals receiving payment penalties in both groups (high volume hospitals: 58.0 versus 51.5%).

Variations in geography did not change the persistence of the disparity (Table 2); in each of the four US Census Bureau regions, racial minority-serving hospitals were more likely to be penalized under the HACRP program. However, regional differences were found in the extent of penalty receipt and corresponding odds ratios. For example, among hospitals in the West, 42.1% of racial minority-serving hospitals were penalized compared to 34.2% of non-racial minority-serving hospitals (absolute difference: +7.9 percentage-points; OR[95%CI]: 1.40[1.27–1.53]). In contrast, in the South, 46.4% of racial minority-serving hospitals were penalized compared to 30.7% of non-racial minority-serving hospitals (absolute difference: +15.8 percentage-points; OR[95%CI]: 1.96[1.92–2.00]).

Discussion

The results of this study reveal that minority-serving hospitals are being disproportionately affected by HACRP payment penalties. Among eight common types of surgical procedures, encompassing >695,000 patients aged ≥65y, receipt of HACRP penalties in FY2017 was approximately 10 percentage-points higher in hospitals managing the highest decile of Black patients. Differences were driven by a stepwise increase in both Domain 1 (patient safety) and Domain 2 (nosocomial infection) HACRP scores that closely aligned with the percentage of Black patients. Nearly identical patterns were observed for variations in the percentage of DSH payments that a hospital received and in the percentage of low-SES patients that a hospital managed. Variations in HACRP penalties again followed suit, demonstrating differential receipt of Medicare payment based on differences in DSH payment receipt and the percentage of low SES patients. Stratified operation-specific analyses pointed toward higher HACRP penalty receipt in six of the eight operations included. Restricted analyses accounting for the influence of teaching status (absolute difference: +13.9 percentage-points) and the severity of patient case-mix (+20.5 percentage-points) both accentuated differences in penalties when limiting hospitals to those at the highest penalty risk (more residents-to-beds, more severe). Restriction to high operative volume, in contrast, reduced the penalty difference (+6.6 percentage-points). Finally, while differences in geography altered the extent of differences in penalty receipt, disparities in HACRP penalties based on race were found in each of the four US Census Bureau regions.

Prior assessment of value-based purchasing policies and other Medicare quality-improvement initiatives suggest that pay-for-performance payment-reduction schemes have the potential to worsen healthcare disparities.^{6–9,12,14–16} The results of this study follow a similar pattern, pointing toward increased rates of HACRP penalization among hospitals managing the largest percentage of Black patients. Such a finding corroborates concerns about potential unintended consequences of HACRP, for while seemingly small in magnitude, the program’s consequential penalty of a 1.0% reduction in all Medicare payments made to a hospital in a given fiscal year¹ could be the difference between a hospital being profitable and operating at a loss, resulting in a reduction of Medicare payments upwards of $1–4 million per year. When combined with the emergence of other Medicare value-based payment programs (e.g. the Hospital Readmissions Reduction Program, Hospital Value-Based Purchasing Program, and ongoing development of alternative payment models) and the increasing percentage of American patients aged ≥65y, the extent of the cumulative penalty on reimbursement is expected to grow. The finding that these penalties fall most heavily on hospitals managing the largest proportion of vulnerable patients is concerning given that many minority-serving hospitals operate with extremely limited resources while managing some of the most medically and socially complex cases.

The results of the study must be interpreted in light of the its limitations. Most come from the study’s reliance on administrative Medicare claims where completeness of information, the potential for absent or misreporting of events, and a lack of nuanced clinical detail can be concerns. The study utilized 100% Medicare fee-for-service data from 2013–2014 linked to Hospital Compare outcomes for the same hospitals in FY2017. While the majority of hospital characteristics are not expected to have changed, it is possible that in relying on 2013–2014 data, the association between hospital characteristics in 2013–2014 (e.g. the percentage of minority-patients) might not be exactly the same as a given hospital’s characteristics when they received a HACRP penalty in FY2017 (2016–2017). What use of Medicare data linked to Hospital Compare outcomes does offer is a national cohort of older adults completely inclusive of the at-risk population in question, enabling meaningful assessment of health policy change. Further studies are warranted to extend these cross-sectional results to look at potential changes in care delivery before and after HACRP implementation and to consider the role that local and regional variations in geography above and beyond those associated with macro-level differences in US Census Bureau regions might play.

Notwithstanding these limitations, our study has important implications. By disproportionately penalizing under-resourced hospitals treating higher-risk patients, the HACRP has the potential to create a ‘reverse Robin Hood effect’^6,7 wherein hospitals with the least resources and greatest need are penalized for having worse outcomes, leaving them with fewer resources to invest in quality improvement and a theoretical progressive decline in the outcomes that they experience. When applied to racial minority-serving, high DSH, and low SES-serving hospitals, such a change has the potential to worsen existing disparities. Restricted analyses looking at differences in teaching status and severity of patient case-mix both showed marked increases in the probability of penalty receipt for racial minority-serving hospitals when they were also larger teaching centers managing a more complex mix of patients. Emerging literature on the HACRP has demonstrated significant differences in penalty receipt attributable to hospital-level differences in teaching status,^3–5,17,18 size,^3,19 cancer center accreditation,²⁰ and safety-net status.²¹ When combined with the finding of more pronounced penalties for minority-serving hospitals with these characteristics, the results of the present study suggest that minority-serving hospitals likely to be the hardest hit by HACRP penalties are the hospitals already at the greatest risk.

As the HACRP continues to develop, efforts are needed to identify and protect patients in vulnerable institutions in order to ensure that disparities do not increase. From a policy perspective, three potential solutions exist. First, like other forms of hospitals managing a disproportionate share of teaching obligations (indirect medical education)²² or Medicaid patients (disproportionate share hospitals),²³ CMS could offer additional compensation to minority-serving hospitals in the form of pre-determined set payments to offset the higher penalty risk. Second, quality metrics used in the HACRP could be designed to account for differences in a hospital’s distribution of patient demographic factors such as race/ethnicity and income. While discussed as a possibility in ongoing reports before Congress,²⁴ such changes have yet to be implemented in quality metrics that CMS actively employs. Third, as suggested by Al Mohajer, Joiner, and Nix,¹⁸ given known differences in hospital-level factors, CMS could stratify hospitals into more homogeneous groups and apply penalties to outlying hospitals within each group. Such an approach is currently being undertaken for dual-eligible patients within the CMS Hospital Readmissions Reduction Program.²⁵ Ultimately, whatever the approach, moving forward careful attention needs to be paid to which hospitals are most affected by HACRP penalties in order to ensure that a well-intended program designed to improve quality and promote value does not lead to an exasperation of disparities among patients most in need. Potential solutions also need to be carefully explored in order to prevent efforts to limit over-penalization from inadvertently leading to unwarranted protection of lower-quality care and unintentional codification of disparities in outcomes among vulnerable patients. Few perfect solutions exist, resulting in a dilemma that remains at the heart of many ongoing policy debates.

The results of this study reveal that racial minority-serving hospitals are being disproportionately affected by HACRP payment penalties. While HACRP scores followed a stepwise increase, disparities in penalty allocation were isolated to hospitals with the largest racial minority-serving extent—a finding which became more pronounced among hospitals with an already heighted risk of penalty receipt. Similar results were observed for differences in hospital DSH payment receipt and the proportion of low SES patients that a hospital treats. Such findings suggest that as the HACRP continues to develop, efforts are needed to identify and protect marginal patients in vulnerable institutions in order to ensure that disparities do not increase. Potential policy solutions include introduction of pre-determined set payments to racial minority-serving hospitals akin to those used for disproportionate share hospitals or indirect medical education, incorporation of demographic information into quality metrics, and hospital stratification prior to assigned penalty receipt.