Abstract
Objective:
Medicare Part D Star Ratings are instrumental in shaping healthcare quality improvement efforts. However, the calculation metrics for medication performance measures for this program have been associated with racial/ethnic disparities. In this study, we aimed to explore whether an alternative program, named Star Plus by us, that included all medication performance measures developed by Pharmacy Quality Alliance and applicable to our study population, would reduce such disparities among Medicare beneficiaries with diabetes, hypertension, and/or hyperlipidemia.
Method:
We conducted an analysis of a 10% random sample of Medicare A/B/D claims linked to the Area Health Resources File. Multivariate logistic regressions with minority dummy variables were used to examine racial/ethnic disparities in measure calculations of Star Ratings and Star Plus, respectively.
Results:
Adjusted results indicated that relative to non-Hispanic Whites (Whites), racial/ethnic minorities had significantly lower odds of being included in the Star Ratings measure calculations: the odds ratios (ORs) for Blacks, Hispanics, Asians, and Others were 0.68 (95% confidence interval [CI]=0.66-0.71), 0.73 (CI=0.69-0.78), 0.88 (CI=0.82-0.93), and 0.92 (CI=0.88-0.97), respectively. In contrast, every beneficiary in the sample was included in Star Plus. Further, racial/ethnic minorities had significantly higher increase in the odds of being included in measure calculation in Star Plus than Star Ratings. The ORs for Blacks, Hispanics, Asians, and Others were 1.47 (CI=1.41-1.52), 1.37 (CI=1.29-1.45), 1.14 (CI=1.07-1.22), and 1.09 (CI=1.03-1.14), respectively.
Conclusions:
Our study demonstrated that racial/ethnic disparities may be eliminated by including additional medication performance measures to Star Ratings.
Keywords: disparities, medication utilization, performance measures, Medicare, ratings
Introduction
Health care in the United States is a paradox: compared with other developed countries, the U.S. spends the most in both aggregated and per-capita terms, but has the lowest life expectancy, highest rate of avoidable deaths, and highest percent of adults with multiple chronic conditions [1]. Most studies have suggested that systemic societal inequality, poor access to care, unhealthy individual behaviors, and asymmetric information are the main culprits of the paradox.[2,3,4] Although it is debatable whether the availability of complete information can help consumers make optimal decisions [5,6], consumers cannot select care most suitable to their needs without adequate information about care quality. To help Medicare beneficiaries make informed choices in their plan enrollment, the Centers for Medicare & Medicaid Services (CMS) established the Part C & D Star Ratings (“Star Ratings” hereafter) program in 2007 [7], one program among a plethora of government-sponsored health care quality reporting programs that have flourished over the past two decades.
Part C refers to Medicare Advantage plans while Part D represents Medicare prescription drug benefits, which may be provided by a Medicare Advantage plan or a standalone prescription drug plan that includes prescription drug coverage (MAPD). All plans under Parts C and D are provided by private companies contracted with CMS [7,8]. A private company may have more than one contract, which in turn may include multiple plans [8]. The Star Ratings program evaluates service quality at the contract level on a five-star scale, with five stars representing the highest quality [9]. For Part D and as of 2019, contracts are assessed by 14 measures and receive a star for each measure as well as for a summary rating, the latter of which is the weighted average of individual measure stars. The ratings are published on the Medicare Plan Finder website, where a low performing icon is given to contracts that have received a summary rating of less than 3 stars for 3 consecutive years [9]. Such contracts are subject to termination [10]. In contrast, high-performing MAPD contracts that have achieved at least 4 stars are rewarded by quality bonus payments [7].
Apart from the incentive mechanism, the Star Ratings program shapes quality improvement efforts through differential weighting in rating calculations. For Part D and as of 2019, a weight of 3 is assigned to each measure of medication adherence for diabetes, hypertension, and hyperlipidemia. By comparison, most other measures are only weighed at 1 or 1.5 each, although the weights of some measures increased in subsequent years (e.g., in 2023, the weight of the measure for complaints about the drug plan was raised to 4) [9,11]. All adherence measures were developed by Pharmacy Quality Alliance (PQA), a nonprofit founded by CMS after the enactment of Medicare Part D in 2006. The organization has evolved from a public-private partnership at its inception to a non-governmental entity [12]. The prominent focus on adherence measures speaks to the severity of nonadherence, which is a medication therapy problem prevalent among all populations, including the older population, and has been associated with worsened health outcomes and high-cost burdens [13,14]. A recent study revealed that the nonadherent proportion of the Medicare population in fee-for-service programs for diabetes, hypertension, and hyperlipidemia medications ranged from 25% to nearly 40% and, if all those nonadherent individuals became adherent, the annual Medicare cost would be reduced by over $20 billion [15].
Given the instrumental role that the adherence measures play in guiding quality improvement efforts, it is imperative to ensure the inclusivity of such measures. However, there have been implications of racial/ethnic disparities whether the measures were examined individually or collectively. For instance, the measure for hypertension medication adherence does not include thiazide-type diuretics and calcium channel blockers, both of which have been identified as initial hypertensive treatment options [16] and reported to be effective among Black patients [17]. Furthermore, a previous study discovered that another adherence measure that targets diabetes medications was found to have excluded 27% of older diabetes patients and the excluded patients were more likely to be racial/ethnic minorities [18]. Among the inclusion criteria for this measure, a patient needs to have at least two prescription refills in the evaluation year [19], but racial/ethnic minority groups have been found to use medications less than White patients [20,21]. Another recent study which utilized a 10% random sample of 2017 Medicare claims data found that, among the elderly with diabetes, hypertension, and/or hyperlipidemia, racial/ethnic minorities were more likely to be excluded from adherence measure calculations. This was true both for beneficiaries with only one disease of interest and among those with more than one disease [22].
Being left out from the calculations of adherence measures means not being part of the target enrollee population for quality improvement initiatives. One possible solution to address the racial/ethnic disparity implications may be to expand the number of medication performance metrics so that the resulting target enrollee population becomes more inclusive. In this study, we created a hypothetical program named Star Plus as an alternative to the Star Ratings medication performance measures. Specifically, Star Plus includes both the existing Star Ratings medication performance metrics and additional PQA-endorsed measures related to medication adherence, use, and safety. We explored whether Star Plus would reduce such racial/ethnic disparities in the inclusion of performance measures among older Medicare beneficiaries with diabetes, hypertension, and hyperlipidemia. We chose the three diseases because they were the conditions targeted in Part D medication performance measures. We hypothesized that, relative to Star Ratings, Star Plus would be associated with lower racial/ethnic disparities in the inclusion of measure calculations.
Methods
Data Sources and Study Sample
This retrospective study was a cross-sectional analysis of beneficiaries in 2017 Medicare databases linked to Area Health Resources File. Medicare databases consisted of the Master Beneficiary Summary File and Medicare Part A, B, and D claim files [23]. The Master Beneficiary Summary File included individuals’ demographic characteristics and enrollment information. Part A and B files provided individuals’ diagnoses and health service utilization records from inpatient and outpatient providers. Part D Event file supplied prescription claim information, including drug name, service date, and days’ supply of medication dispensed [23]. The Area Health Resources File provided information on the socioeconomic characteristics of each beneficiary’s residence at the county level [24].
The study sample was obtained using a 10% systematic random sampling of fee-for-service Medicare beneficiaries enrolled in 2017. To reduce heterogeneity, included individuals were required to meet the following criteria: aged at least 65 years old, alive at the end of 2017, and had continuous Parts A, B, and D coverage throughout 2017. In addition, those who did not have prescription claims for medications to treat at least one of the three disease states of interest (diabetes, hypertension, and hyperlipidemia) were excluded from the final study sample.
Medication Performance Measures in Star Ratings and Star Plus
Star Ratings Measures
The 2019 measures were used because the measurement year of interest was 2017. The apparent two-year lag was due to a gap in the timeline between when data was generated and when the ratings based on those data were determined [9]. For instance, data generated by a provider in the measurement year 2019 was collected and evaluated in 2020, and the results were used to determine the provider’s Star Ratings for 2021.
The 2019 Star Ratings Part D medication performance measures included four PQA-endorsed metrics: statins use in persons with diabetes, diabetes medication adherence, hypertension medication adherence, and hyperlipidemia medication adherence [9]. The PQA developed these measures by collaborating with various stakeholders, including PQA members and external panels [12]. Specifically, in the measure developing process, each measure must go through the following steps: 1) measure conceptualization, 2) measure specification, 3) measure testing, 4) measure endorsement, and 5) implementation and maintenance phases. Throughout the above cycle of the measure development, the PQA evaluates measures using the following measure evaluation criteria aligned with the Blueprint for the CMS Measure Management System: importance, scientific acceptability, feasibility, and usability [12]. We followed the PQA technical specifications when applying inclusion/exclusion criteria for each metric. All four metrics had the following inclusion criteria: had at least two prescription fills of eligible medication on different dates; did not have an end stage renal disease diagnosis; and were not in hospice care. Each adherence metric had a criterion requiring that the first eligible prescription fill must have occurred at least 91 days before the end of the year. The statins use in persons with diabetes measure’s only additional criterion was an age limit of up to 75 years. Eligible prescription medications were the following therapeutic categories: diabetic medications for the diabetes medication adherence and statins use in persons with diabetes measures, statin medications for the hyperlipidemia medication adherence measure, and renin angiotensin receptor antagonists for the hypertension medication adherence measure. Additionally, the diabetes medication adherence measure excluded individuals with insulin prescription claims, while the hypertension medication adherence measure excluded those with claims for the combination of sacubitril and valsartan [19].
Star Plus Measures
The hypothetical Star Plus program included all four Star Ratings measures plus all PQA-endorsed medication utilization measures applicable to the target study population, resulting in a total of 25 measures. The PQA list of 25 measures used for creating the hypothetical program can be found in Table 1. A few PQA-endorsed medication utilization measures were not applicable to the study population. As with the four Star Ratings measures, each of the additional metrics had its own inclusion/exclusion criteria. Likewise, we followed the PQA technical specifications when applying the criteria for each. Among the additional measures, the use of benzodiazepine sedative hypnotics medications in the older population had the least restrictive inclusion criteria and no exclusion criteria: individuals were included as long as they were at least 65 years of age and had continuous enrollment during the measurement year [19].
Table 1.
Pharmacy Quality Alliance Medication Utilization Measures Included in Star Plus
| PQA Measure | Measures |
|---|---|
| Adherence Measures | |
| Adherence to diabetic medications* | |
| Adherence to renin angiotensin system antagonists (hypertension medications)* | |
| Adherence to statins (hyperlipidemia medications)* | |
| Adherence to beta-blockers | |
| Adherence to calcium channel blockers | |
| Adherence to biguanides | |
| Adherence to dipeptidyl peptidase-4 inhibitors | |
| Adherence to sulfonylureas | |
| Adherence to thiazolidinediones | |
| Adherence to non-warfarin oral anticoagulants | |
| Adherence to long-acting inhaled bronchodilator agents in treating chronic obstructive pulmonary disease | |
| Adherence to antiretroviral medications | |
| Adherence to non-fused disease-modifying agents used to treat multiple sclerosis | |
| Appropriateness | |
| Statin use in persons with diabetes* | |
| Inappropriate use of diabetes medications at higher than daily recommended dosage | |
| Suboptimal control in medication therapy for persons with asthma | |
| Suboptimal cholesterol management in coronary artery disease | |
| Safety | |
| Concurrent use of medications that leads to drug-drug interactions | |
| Use of antipsychotic medication in persons with dementia | |
| Use of high-risk medications in the elderly | |
| Use of benzodiazepine sedative hypnotics in the elderly | |
| Use of multiple anticholinergic medications in older adults | |
| Use of multiple central-nervous system active medications in older adults | |
| Concurrent use of opioids and benzodiazepines | |
| Use of opioids at high dosage in persons without cancer |
Note:
indicates measures included in the 2019 Star Ratings Part D medication performance measures.
Outcome Measures
Two binary outcome variables were constructed to examine the likelihood of inclusion in measure calculations. One outcome variable represented the likelihood of being included in the measure calculations of Star Ratings. If all inclusion/exclusion criteria were met for at least one of the four 2019 Star Ratings Part D medication performance measures, the individual was coded as one (included in at least one Star Ratings measure = 1; not included in any of the Star Ratings measures = 0). The other outcome variable measured the likelihood of inclusion in the measure calculations of Star Plus but not Star Ratings. Since Star Plus encompassed all Star Ratings measures, inclusion in Star Plus but not Star Ratings was equivalent to inclusion in one of the additional PQA-endorsed metrics in Star Plus. Individuals were coded as one if they met the inclusion criteria for at least one of the additional metrics in Star Plus (included in Star Plus but not Star Ratings = 1; included in both Star Ratings and Star Plus = 0).
Theoretical Framework
The Gelberg-Andersen’s Behavioral Model for Vulnerable Populations was employed to guide the selection of explanatory variables because the outcomes were associated with medication utilization [25]. Predisposing variables that influence medication utilization included age, gender, race/ethnicity, the proportion of married-couple families, the proportion of people with at least high school education, income per capita, and the proportion of people without health insurance. Enabling factors that facilitate the use of medication included indicator variables for Medicaid eligibility, metropolitan statistical area, Health Professional Shortage Area, and census regions. Need factor, representing the necessity for medications, included a CMS risk adjustment summary score that was determined based on the demographic characteristics and disease diagnosis records of a beneficiary; the risk score is obtained by adding coefficients associated with disease factors and demographics of a beneficiary, and the coefficients were estimated by CMS based on fee-for-service data [26]. Because the risk adjustment summary score measured the financial risks that a beneficiary posed to plan providers, it served as a proxy for beneficiary health status. Age, gender, race/ethnicity, and risk adjustment summary scores were individual-level characteristics obtained from the Medicare databases. All other covariates were county-level and collected from the Area Health Resources File.
Statistical Analysis
We categorized race/ethnicity into five groups: non-Hispanic Whites (Whites), African Americans (Blacks), Hispanics, Asians and Pacific Islanders (Asians), and all other races/ethnicities (Others). Descriptive statistics were first obtained, and then bivariate analyses with Chi-squared tests were performed to compare the proportions of included individuals across groups by outcome measure.
Racial/ethnic disparities were examined by comparing the inclusion outcomes for Whites with the outcomes for the minority groups. Specifically, a two-stage analysis was conducted using multivariate logistic regression models that included dummy variables for minority group, with Whites being the reference group. In the first stage, we examined the outcome, the likelihood of being included in the calculations of Star Ratings metrics, to determine whether disparities existed in the Star Ratings measure calculations. An odds ratio (OR) for a minority indicator variable of less than one would suggest that the minority group was less likely than Whites to be included in the calculations of the Star Ratings measures. In the second stage, we tested whether Star Plus was associated with decreased racial/ethnic disparities relative to Star Ratings. The likelihood of being included in Star Plus but not Star Ratings was compared among racial/ethnic groups. A larger-than-one OR for a minority dummy would imply that the minority racial/ethnic group would be more likely to be included in the calculations of the additional metrics in Star Plus.
In each stage of the analysis, regressions were run on the same need-based and demand-based models, both of which were grounded in the Gelberg-Andersen’s model with a respective focus on factors that affect health care need and demand [27]. The explanatory variables in the need-based model included age, gender, race/ethnicity, and risk adjustment summary score, while those in the demand-based model included all the factors in the Gelberg-Andersen’s model. In all models, standard errors were clustered at the county level to account for possible correlation of errors within the same county. All analyses were performed with SAS Enterprise 7.1 (Cary, NC) at the CMS Virtual Research Data Center. The Institutional Review Board at the corresponding author’s institution deemed this study exempt (approval number: #20-07197-XM). Informed consent is not applicable for this study.
Results
The final study sample consisted of 1,207,535 Medicare beneficiaries, including 81.82% Whites, 7.09% Blacks, 5.33% Hispanics, 3.17% Asians, and 2.59% Others. All characteristics were generally significantly different across racial and ethnic groups (p < .01) (Table 2). Results from the unadjusted comparison show that, for example, among predisposing factors, Whites were older than Blacks and Others. Compared to minority racial/ethnic groups, Whites overall were more likely to reside in counties with higher proportions of married-couple families and people with at least high school degree. Compared to Blacks and Hispanics, Whites were more likely to live in counties with lower proportions of uninsured people. Among enabling factors, higher proportions of Whites had Medicaid in general and resided in non-metropolitan statistical area and non-health professional shortage area. In terms of the need factor, Whites had lower Risk Adjustment Summary Score than Blacks and Hispanics.
Table 2.
Beneficiary Characteristics across Racial and Ethnic Groups
| Characteristics | Non-Hispanic Whites (n= 988 003, 81.82%) |
Blacks (n=85 593, 7.09%) |
Hispanics (n=64 405, 5.33%) |
Asians/Pacific Islanders (n=38 225, 3.17%) |
Others (n=31 309, 2.59%) |
|||||
|---|---|---|---|---|---|---|---|---|---|---|
| Number | % | Number | % | Number | % | Number | % | Number | % | |
| Predisposing Factors | ||||||||||
| Age, mean (SD) | 75.97 (7.32) |
75.26* (7.37) |
75.70 (7.34) |
76.53* (7.48) |
71.94* (5.89) |
|||||
| Male | 423 670 | 42.88 | 29 527* | 34.50 | 24 980* | 38.79 | 15 969* | 41.78 | 18 092* | 57.79 |
| Proportion Married-Couple Families, mean (SD)a | 0.74 (0.06) |
0.67* (0.08) |
0.71* (0.06) |
0.72* (0.06) |
0.73* (0.07) |
|||||
| Proportion Education >=High School, mean (SD)a | 0.88 (0.05) |
0.86* (0.05) |
0.82* (0.08) |
0.86 (0.05) |
0.88 (0.05) |
|||||
| Income per Capita (in $1000), mean (SD)a | 50.44 (16.52) |
50.01 (17.15) |
51.83 (20.53) |
63.87* (22.35) |
53.98* (19.29) |
|||||
| Proportion No Insurance, mean (SD)a | 0.10 (0.05) |
0.11* (0.04) |
0.13* (0.07) |
0.09* (0.04) |
0.10 (0.05) |
|||||
| Enabling Factors | ||||||||||
| Medicaid | 115 942 | 11.73 | 34 584* | 40.41 | 37 181* | 57.73 | 21 354* | 55.86 | 6 570* | 20.98 |
| Metropolitan Statistical Areaa | 757 633 | 76.68 | 72 211* | 84.37 | 57 734* | 89.64 | 36 935* | 96.63 | 25 116* | 80.22 |
| Health Professional Shortage Areaa | 888 914 | 89.97 | 81 087* | 94.74 | 61 846* | 96.03 | 35 900 | 93.92 | 28 792* | 91.96 |
| Census Regionsa | ||||||||||
| Northeast | 200 621 | 20.31 | 13 536* | 15.81 | 10 745* | 16.68 | 6 937* | 18.15 | 7 147* | 22.83 |
| Midwest | 256 523 | 25.96 | 16 095 | 18.80 | 5 945 | 9.23 | 3 602 | 9.42 | 6 584 | 21.03 |
| South | 373 621 | 37.82 | 50 787 | 59.34 | 23 359 | 36.27 | 7 547 | 19.74 | 9 219 | 29.45 |
| West | 157 238 | 15.91 | 5 175 | 6.05 | 24 356 | 37.82 | 20 139 | 52.69 | 8 359 | 26.70 |
| Need Factor | ||||||||||
| Risk Adjustment Summary Score, mean (SD) | 1.35 (1.20) |
1.74* (1.47) |
1.64* (1.36) |
1.37 (1.14) |
1.15* (1.10) |
|||||
| Inclusion Outcomes | ||||||||||
| Included in SR | 934 299 | 94.56 | 77 962* | 91.08 | 58 840* | 91.36 | 35 609* | 93.16 | 29 588 | 94.50 |
| Included in SP | 988 003 | 100 | 85 593 | 100 | 64 405 | 100 | 38 225 | 100 | 31 309 | 100 |
| Not Included in SR but in SP | 53 704 | 5.44 | 7 631* | 8.92 | 5 565* | 8.64 | 2 616* | 6.84 | 1 721 | 5.50 |
Note:
Indicates a county-level characteristic.
Indicates characteristic was different from non-Hispanic Whites by unadjusted pairwise comparison (p < 0.05); significance for census regions is for overall distribution, but not regions-specific. Abbreviations: SD=standard deviation; SR = Star Ratings; SP = Star Plus.
Bivariate analysis results revealed the outcome measure of inclusion in Star Ratings and Star Plus (Table 2). The proportions of Whites, Blacks, Hispanics, Asians, and Others included in the calculation of Star Ratings measures was 94.56%, 91.08%, 91.36%, 93.16%, and 94.50%, respectively. Significant racial/ethnic disparities were observed in the inclusion outcome related to Star Ratings measures except Others. The proportion of Whites included in the calculation of at least one Star Ratings measure was 3.48 percentage points higher than Blacks, 3.20 percentage points higher than Hispanics, and 1.40 percentage points higher than Asians. In contrast, significant reduction in racial/ethnic disparities were found in the inclusion outcome related to Star Plus measures. In fact, the magnitude of increase in the included proportion for each racial/ethnic group suggests that Star Plus picked up all the beneficiaries excluded from Star Ratings. In other words, all individuals in the study sample were included in Star Plus.
The patterns of racial/ethnic disparities associated with Star Ratings measure calculations after adjusting for need-based and demand-based factors are presented in Table 3. The results were consistent with the findings from bivariate analyses. After adjusting for need-based factors, racial/ethnic minorities had considerably lower odds of being included in the Star Ratings measure calculations than Whites. The ORs for Blacks, Hispanics, Asians, and Others were, respectively, 0.66 (95% confidence interval or CI=0.64-0.68), 0.66 (CI=0.63-0.69), 0.79 (CI=0.74-0.84), and 0.89 (CI=0.84-0.93). After adjusting for demand-based factors, the ORs for racial/ethnic minorities increased slightly. For instance, the ORs was 0.73 (CI=0.69-0.78) for Hispanics and was 0.88 (CI=0.82-0.93) for Asians.
Table 3.
Odds of Inclusion in the Calculations of Star Ratings Measures
| Characteristics | Demand-Based | Need-Based | ||
|---|---|---|---|---|
| OR | 95% CI | OR | 95% CI | |
| Predisposing Factors | ||||
| Age | 0.990 | 0.989-0.992 | 0.991 | 0.990-0.992 |
| Male | 1.09 | 1.07-1.12 | 1.11 | 1.08-1.13 |
| Race/Ethnicity | ||||
| Blacks | 0.68 | 0.66-0.71 | 0.66 | 0.64-0.68 |
| Hispanics | 0.73 | 0.69-0.78 | 0.66 | 0.63-0.69 |
| Asians/Pacific Islanders | 0.88 | 0.82-0.93 | 0.79 | 0.74-0.84 |
| Others | 0.92 | 0.88-0.97 | 0.89 | 0.84-0.93 |
| Proportion Married-couple Families* | 1.06 | 0.88-1.29 | ||
| Proportion Education >= High School* | 0.38 | 0.29-0.49 | ||
| Income per Capita (in $1000)* | 1.0003 | 0.9995-1.0011 | ||
| Proportion No Insurance* | 0.25 | 0.19-0.34 | ||
| Enabling Factors | ||||
| Medicaid | 0.85 | 0.83-0.88 | ||
| Metropolitan Statistical Area* | 0.97 | 0.95-0.99 | ||
| Health Professional Shortage Area* | 0.98 | 0.95-1.01 | ||
| Census Regions* | ||||
| Midwest | 0.95 | 0.92-0.98 | ||
| South | 0.92 | 0.89-0.96 | ||
| West | 0.80 | 0.77-0.83 | ||
| Need Factor | ||||
| Risk Adjustment Summary Score | 0.784 | 0.779-0.788 | 0.777 | 0.772-0.781 |
Note:
indicates a county-level characteristic. Reference groups: female, non-Hispanic Whites, non-metropolitan statistical area, non-health professional shortage area, and Northeast region.
Abbreviations: OR=odds ratio; CI=confidence interval.
The results also revealed that some beneficiary characteristics were significantly associated with the inclusion outcome (Table 3). Specifically, being male (OR=1.09; CI=1.07-1.12) was associated with a higher likelihood of inclusion in Star Ratings calculations. On the contrary, being older (OR=0.990; CI=0.989-0.992), living in counties with a higher proportion of people with at least high school degree (OR=0.38; CI=0.29-0.49), living in counties with a higher proportion of people without health insurance (OR=0.25; CI=0.19-0.34), with Medicaid insurance (OR=0.85; CI=0.83-0.88), living in a metropolitan statistical area (OR=0.97; CI=0.95-0.99), and having a higher risk adjustment summary score (OR=0.784; CI=0.779-0.788) were associated with a lower likelihood of inclusion. Likewise, relative to those living in the Northeast census region, beneficiaries who resided in the Midwest (OR=0.95; CI=0.92-0.98), the South (OR=0.92; CI=0.89-0.96), and the West (OR=0.80; CI=0.77-0.83) were less likely to be included.
Table 4 presents results from the adjusted analyses of racial/ethnic disparity associated with the calculations of Star Plus but not Star Ratings measures. The binary outcome variable for this set of analyses was the likelihood of being included in the measure calculations of Star Plus but not Star Ratings. The results reinforced the findings from bivariate analyses. After adjusting for need-based factors and compared with Whites, racial/ethnic minorities experienced higher increase in odds of being included in measure calculations under Star Plus compared to Star Ratings. The ORs for Blacks, Hispanics, Asians, and Others were, respectively, 1.52 (CI=1.48-1.57), 1.52 (CI=1.44-1.59), 1.27 (CI=1.20-1.35), and 1.13 (CI=1.07-1.19). After adjusting for demand-based factors, the ORs for racial/ethnic minorities decreased slightly. For instance, the odds of being included for Hispanics was 1.37 times that of Whites (CI=1.29-1.45). Note that all results presented in Table 4 are the inverse of the results presented in Table 3.
Table 4.
Odds of Inclusion in the Calculations of Star Plus but not Star Ratings Measures
| Characteristics | Demand-Based | Need-Based | ||
|---|---|---|---|---|
| OR | 95% CI | OR | 95% CI | |
| Predisposing Factors | ||||
| Age | 1.010 | 1.008-1.011 | 1.009 | 1.008-1.011 |
| Male | 0.91 | 0.90-0.93 | 0.91 | 0.89-0.92 |
| Race/Ethnicity | ||||
| Blacks | 1.47 | 1.41-1.52 | 1.52 | 1.48-1.57 |
| Hispanics | 1.37 | 1.29-1.45 | 1.52 | 1.44-1.59 |
| Asians/Pacific Islanders | 1.14 | 1.07-1.22 | 1.27 | 1.20-1.35 |
| Others | 1.09 | 1.03-1.14 | 1.13 | 1.07-1.19 |
| Proportion Married-couple Families* | 0.94 | 0.77-1.14 | ||
| Proportion Education >= High School* | 2.67 | 2.04-3.49 | ||
| Income per Capita (in $1000)* | 1.000 | 0.999-1.001 | ||
| Proportion No Insurance* | 4.00 | 2.99-5.37 | ||
| Enabling Factors | ||||
| Medicaid | 1.17 | 1.13-1.21 | ||
| Metropolitan Statistical Area* | 1.03 | 1.01-1.06 | ||
| Health Professional Shortage Area* | 1.02 | 0.99-1.05 | ||
| Census Regions* | ||||
| Midwest | 1.05 | 1.02-1.09 | ||
| South | 1.08 | 1.05-1.12 | ||
| West | 1.25 | 1.21-1.30 | ||
| Need Factor | ||||
| Risk Adjustment Summary Score | 1.276 | 1.269-1.284 | 1.29 | 1.28-1.30 |
Note:
indicates a county-level characteristic. Reference groups: female, non-Hispanic Whites, non-metropolitan statistical area, non-health professional shortage area, and Northeast region.
Abbreviations: OR=odds ratio; CI=confidence interval.
Discussion
This study used the most recent Medicare claims data to explore the racial/ethnic disparities reduction prospect of Star Plus among older Medicare beneficiaries with diabetes, hypertension, and/or hyperlipidemia. Star Plus is a hypothetical program this study utilized as an alternative to the 2019 Part D Star Ratings medication performance measures. This study suggested that racial/ethnic disparities in the inclusion criteria under Star Ratings were eliminated by using Star Plus. The result was achieved by expanding the performance measures utilized and thus increasing the pool of beneficiaries that met all the inclusion criteria for at least one of the additional measures in consideration. Given that the metric for benzodiazepine sedative hypnotics medications use had the most lenient inclusion criteria among all additional measures, this metric played the pivotal role in including the entire study population in the hypothetical Star Plus program.
In addition, findings from the adjusted analysis of the inclusion patterns of the current Part D Star Ratings metrics were consistent with the results of an earlier study [22]. While the earlier study focused on adherence measures only and on disease-specific disparities [22], the present study collectively examined all medication performance measures in Star Ratings, including adherence metrics for diabetes, hypertension, and hyperlipidemia, as well as the utilization measure for statin use in persons with diabetes. This study further confirmed that relative to Whites, racial/ethnic minorities had significantly lower odds of being included in the Star Ratings measure calculations. Such results were robust against the use of different sets of covariates from demand- and need-based models. These results demonstrated that racial/ethnic disparities persisted, whether only a subset or all of the medication performance measures were examined.
Ensuring equitable care was one of the six aims for health care quality improvement highlighted by the Institute of Medicine two decades ago [2]; yet earlier studies and our present study revealed that racial/ethnic disparities exist in the inclusion criteria of Star Ratings medication performance measures. These are not the only disparity implications that have been associated with Star Ratings. Another similar type of disparity has been identified in the correlation between plan ratings and beneficiary sociodemographic status, such as race/ethnicity and income. Contracts with a higher proportion of minority and low-income enrollees were found to be more likely to receive lower ratings [28-31]. Due to this disparity implication, some have suggested that ratings be adjusted for patient sociodemographic status so that contracts do not avoid high-risk patients [29-31]. The suggestion was endorsed by the National Quality Forum, which regarded such risk adjustments as crucial for ensuring a comparison of “apples to apples” [32]. CMS responded to this suggestion by implementing a Categorical Adjustment Index starting from 2017 [33]. The Categorical Adjustment Index adjusts for the proportion of enrollees who receive low-income subsidy, have disabilities, or are dually eligible for Medicare and Medicaid [33]. However, this is only an interim policy while the measure developers review the metrics, and the Categorical Adjustment Index calculation is only based on a subset of the Star Ratings measures [9]. For instance, the 2019 Categorical Adjustment Index calculation included merely one adherence measure, hypertension medication adherence [9].
The risk adjustment solution has triggered a “nature vs. nurture” debate over the extent to which health care providers are responsible for their patients’ health behaviors [28]. Proponents of the “nature” perspective believe that risk adjustment helps level the playing field for quality performance evaluation. By doing so, the adjustment implicitly acknowledges that patients’ health behaviors are inherently different due to their sociodemographic status. For instance, if minorities are found more likely to be nonadherent to diabetes medications, it is because of cultural beliefs and/or other environmental factors of the minority group. This suggests that these intrinsic factors are beyond a health care provider’s responsibility and thus is unfair for such factors to skew the providers’ performance scores. Proponents of the “nurture” perspective, however, maintain that patients’ health behaviors can be changed. This view supports that providers are responsible for catalyzing desirable changes in the high-risk behaviors if such behaviors are assessed in performance metrics. The very essence of quality assessment metrics is to motivate providers to understand better the patients they serve and devise effective ways to improve suboptimal health behaviors [28].
While the “nurture” view has been primarily centered around the disparity implications related to the association between ratings and patient sociodemographic status, it aligns with the premise of our research. Providers’ quality improvement efforts can change medication use and adherence behaviors, and the Star Ratings metrics should guide such efforts. The current momentum of incorporating risk adjustment in Star Ratings should not obscure the equally, if not more, important implication of racial/ethnic disparities in measure calculations. Risk adjustment may result in more plan options for minorities since contracts will be less likely to shun high-risk patients. However, improving the enrollment of minorities in plans does not equate to improving the inclusion of minorities in the population targeted for quality improvement. For example, suppose minorities are known to be less likely to be included in quality improvement metrics in the first place, then minorities will be less likely to be considered as a high priority when providers strive to improve their metric performance scores. Thus, racial/ethnic disparities in performance measure calculations would likely continue despite offering minorities more plan enrollment options.
Our analyses have demonstrated that Star Plus eliminated racial/ethnic disparities associated with inclusion in medication performance measures. While it may not be feasible for CMS to adopt Star Plus in its entirety, which encompasses 25 PQA-endorsed medication performance measures, the finding offers a conceptual direction for solutions to racial/ethnic disparities in performance measures and programs. As CMS stated, changes to measure specifications can only be made by the measure developer, and the endorsement process takes many years [34]. Therefore, expanding the number of metrics may serve as an interim solution, similarly as the Categorical Adjustment Index has, while measure specifications undergo a thorough review. Determining exactly how many and which additional measures should be included in the Star Ratings program should be a decision derived from engaging CMS’ key stakeholders such as contracted providers, measure developers, patient advocacy groups, and the research communities.
This study had a few limitations. First, the primary data used were Medicare claims, which only provide a limited number of individual-level variables. Even though we supplemented the data with county-level information obtained from Area Health Resources File, county characteristics are not precise proxies for individual characteristics. Second, our proposed program did not intend to address the disparity implications in the measure specifications of Star Ratings because the Star Ratings measures were developed systematically following a widely accepted procedure. Therefore, the disparity implications for each measure remain unchanged in our study, although expanding the pool of Star Ratings measures in Star Plus can increase the number of minority groups covered by Part D quality assessment. Third, the strategy of adding more measures may carry unintended downsides. For instance, while informative, the proposed new set of many measures can bring about a need to educate patients on all measures to avoid information overload. Further, some plans may not be able to calculate certain measures due to a small number of patients meeting the inclusion criteria. Finally, we did not differentiate beneficiaries by their number of conditions. Since the metrics in both Star Ratings and Star Plus are disease-specific, theoretically the more conditions that one had, the higher the chance to be included in measure calculations. The disparities patterns might be different if they were stratified by the number of conditions and thus might be more informative for policy efforts attempting to target a particular segment of beneficiary populations. However, dissecting disparities by the degree of disease severity was not the focus of this study, which was intended to present a comprehensive picture of disparity implications associated with Star Ratings Part D medication performance measures as a whole.
Conclusion
Quality measurement underpins quality improvement. Specifically, measurement sheds light on the status quo, and the definition of how quality is assessed shapes the direction of quality improvement initiatives. Medicare Star Ratings is a national program that evaluates the performance of contracted private companies providing care to millions of Medicare beneficiaries; thus, it is of paramount importance that the process of measuring quality is inclusive so that disadvantaged populations such as racial/ethnic minorities are not unintentionally left out. This study presented a conceptual alternative to Star Ratings and demonstrated that racial/ethnic disparities in inclusion in Star Ratings measures may be reduced or even eliminated by adding additional medication utilization measures with broader inclusion criteria. Further research is needed to explore the specifics of the alternative performance and quality assessment method, such as different combinations of additional metrics, to better inform the continued development of the Star Ratings program.
Acknowledgements
The authors would like to acknowledge assistance with the formatting of the final manuscript from Lorraine Todor and Hannah Foster, doctor of pharmacy students at the University of Tennessee Health Science Center College of Pharmacy.
Declaration of funding
Research reported in this publication was supported by the National Institute on Aging of the National Institutes of Health under Award Number R01AG049696. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
Declaration of financial and other relationships
Research reported in this publication was supported by the National Institute on Aging of the National Institutes of Health under Award Number R01AG049696. Xiaobei Dong, Chi Chun Steve Tsang, Jamie A. Browning, Jim Y. Wan, Samuel Dagogo-Jack, and Yongbo Sim: None. Marie A. Chisholm-Burns: Received funding from Carlos and Marguerite Mason Trust. William C. Cushman: Received grant funding from Eli Lilly. Junling Wang: Received funding from AbbVie, Curo, Bristol Myers Squibb, Pfizer, and Pharmaceutical Research and Manufacturers of America (PhRMA) and serves on Value Assessment & Health Outcomes Research Advisory Committee of the PhRMA Foundation.
Peer reviewers on this manuscript have no relevant financial or other relationships to disclose.
Data Availability Statement
Medicare databases are United States federal databases sponsored by the Centers for Medicare & Medicaid Services (CMS). These data are available to researchers through the Research Data Assistance Center (ResDAC) at the University of Minnesota, according to a strict protocol for data requests. Users of Medicare databases cannot disclose to, nor share the data with, individuals not listed in the Data Use Agreement. ResDAC can be reached via email at resdac@umn.edu, or by phone at 888-973-7322.
References
- [1].Tikkanen R. The CommonWealth Fund. Multinational comparisons of health systems data, 2019. [Cited 2022 Dec 7]. Available from: https://www.commonwealthfund.org/sites/default/files/2020-01/Tikkanen_multinational_comparisons_hlt_sys_data_2019_01-30-2020.pdf. [Google Scholar]
- [2].Institute of Medicine Committee on Quality of Health Care in America. Crossing the quality chasm: a new health system for the 21st century. Washington (DC): National Academies Press. 2001; doi: 10.17226/10027 [DOI] [Google Scholar]
- [3].Avendano M, Kawachi I. Why do Americans have shorter life expectancy and worse health than do people in other high-income countries?. Annu Rev Public Health. 2014;35:307–325. doi: 10.1146/annurev-publhealth-032013-182411 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [4].Konetzka RT, Grabowski DC, Perraillon MC, et al. Nursing home 5-Star Rating system exacerbates disparities in quality, by payer source. Health Aff (Millwood). 2015; 34(5):819–827. doi: 10.1377/hlthaff.2014.1084 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [5].McWilliams JM, Afendulis CC, McGuire TG, Landon BE. Complex Medicare advantage choices may overwhelm seniors--especially those with impaired decision making. Health Aff (Millwood). 2011;30(9):1786–1794. doi: 10.1377/hlthaff.2011.0132 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [6].Reid RO, Deb P, Howell BL, Shrank WH. Association between Medicare Advantage plan star ratings and enrollment. JAMA. 2013;309(3):267–274. doi: 10.1001/jama.2012.173925 [DOI] [PubMed] [Google Scholar]
- [7].Centers for Medicare & Medicaid Services. Part C and D performance data. Updated Oct 2022 [Cited 2022 Dec 7]. Available from: https://www.cms.gov/medicare/prescription-drug-coverage/prescriptiondrugcovgenin/performancedata.html.
- [8].Frakt AB, Pizer SD. A first look at the new Medicare prescription drug plans. Health Aff (Millwood). 2006; 25:w252–w261. doi: 10.1377/hlthaff.25.w252 [DOI] [PubMed] [Google Scholar]
- [9].Centers for Medicare & Medicaid Services. Medicare 2019 Part C & D Star Ratings technical notes. Updated March 2019 [Cited 2022 Dec 7]. Available from: https://www.cms.gov/Medicare/Prescription-Drug-Coverage/PrescriptionDrugCovGenIn/Downloads/2019-Technical-Notes.pdf.
- [10].Centers for Medicare & Medicaid Services. Medicare program; changes to the Medicare Advantage and the Medicare prescription drug benefit programs for contract year 2013 and other changes. 2012. [Cited 2022 Dec 7]. Available from: https://www.federalregister.gov/documents/2012/04/12/2012-8071/medicare-program-changes-to-the-medicare-advantage-and-the-medicare-prescription-drug-benefit.
- [11].Centers for Medicare & Medicaid Services. Medicare 2023 Part C & D Star Ratings technical notes. Updated January 2023 [Cited 2023 Feb 8]. Available from: https://www.cms.gov/files/document/2023-star-ratings-technical-notes.pdf
- [12].Pharmacy Quality Alliance. About PQA. [Cited 2022 Dec 7]. Available from:https://www.pqaalliance.org/our-story.
- [13].Osterberg L, Blaschke T. Adherence to medication. N Engl J Med. 2005; 353:487–497. doi: 10.1056/NEJMra050100 [DOI] [PubMed] [Google Scholar]
- [14].Gellad WF, Grenard J, McGlynn EA. A review of barriers to medication adherence: A framework for driving policy options. Santa Monica (CA): RAND Corporation. 2009. [Cited 2022 Dec 7]. Available from: https://www.rand.org/pubs/technical_reports/TR765.html. [Google Scholar]
- [15].Lloyd JT, Maresh S, Powers CA, et al. How much does medication nonadherence cost the Medicare fee-for-service program? Med Care. 2019; 57:218–224. doi: 10.1097/MLR.0000000000001067 [DOI] [PubMed] [Google Scholar]
- [16].Ta JT, Erickson SC, Qiu WA, et al. Is there a relationship between Part D medication adherence and Part C intermediate outcomes Star Ratings measures? J Manag Care Spec Pharm. 2016; 22:787–795. doi: 10.18553/jmcp.2016.22.7.787 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [17].Egan BE. Treatment of hypertension in Black individuals. In: Up To Date. 2017. [Cited 2022 Dec 7]. Available from: https://www.uptodate.com/contents/treatment-of-hypertension-in-black-patients/print. [Google Scholar]
- [18].Schmittdiel J, Raebel M, Dyer W, et al. Medicare Star excludes diabetes patients with poor CVD risk factor control. Am J Manag Care. 2014; 20:e573–e581. [PMC free article] [PubMed] [Google Scholar]
- [19].Pharmacy Quality Alliance. Technical specifications for PQA-endorsed measures. Alexandria. 2018; vol. A. [Google Scholar]
- [20].Briesacher B, Limcangco R, Gaskin D. Racial and ethnic disparities in prescription coverage and medication use. Health Care Financ Rev. 2003; 25:63–76. [PMC free article] [PubMed] [Google Scholar]
- [21].Wang J, Zuckerman IH, Miller NA, et al. Utilizing new prescription drugs: disparities among non-Hispanic Whites, non-Hispanic Blacks, and Hispanic Whites. Health Serv Res. 2007; 42:1499–1519. doi: 10.1111/j.1475-6773.2006.00682.x [DOI] [PMC free article] [PubMed] [Google Scholar]
- [22].Tsang CCS, Wan JY, Chisholm-Burns MA, et al. Racial/ethnic disparities in measure calculations for Part D Star Ratings among Medicare beneficiaries with diabetes, hypertension, and/or hyperlipidemia. Res Social Adm Pharm. 2021; 17:1469–1477. doi: 10.1016/j.sapharm.2020.11.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [23].Research Data Assistance Center. Data File Search. [Cited 2022 Dec 7]. Available from: https://www.resdac.org/cms-data?tid_1%5B%5D=1.
- [24].Health Resources & Services Administration. Area Health Resources Files. [Cited 2022 Dec 7]. Available from: https://data.hrsa.gov/topics/health-workforce/ahrf.
- [25].Gelberg L, Andersen RM, Leake BD. The Behavioral Model for Vulnerable Populations: application to medical care use and outcomes for homeless people. Health Serv Res. 2000; 34:1273–1302. [PMC free article] [PubMed] [Google Scholar]
- [26].Centers for Medicare & Medicaid Services. Part D payment and risk adjustment. Updated Dec 2021 [Cited 2022 Dec 7]. Available from: https://www.cms.gov/Medicare/Prescription-Drug-Coverage/DrugCoverageClaimsData/RxClaims_PaymentRiskAdjustment.
- [27].Escarce JJ, Kapur K. Racial and ethnic differences in public and private medical care expenditures among aged Medicare beneficiaries. Milbank Q. 2003; 81:249–75, 172. doi: 10.1111/1468-0009.t01-1-00053 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [28].Desai V, Nau D, Conklin M, et al. Impact of environmental factors on differences in quality of medication use: an insight for the Medicare Star Rating system. J Manag Care Spec Pharm. 2016; 22:779–86. doi: 10.18553/jmcp.2016.22.7.779 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [29].Dharmarajan S, Bentley JP, Banahan BF III, et al. Measuring pharmacy performance in the area of medication adherence: addressing the issue of risk adjustment. J Manag Care Spec Pharm. 2014; 20:1057–68. doi: 10.18553/jmcp.2014.20.10.1057.26 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [30].Sorbero ME, Paddock SM, Damberg CL, et al. Adjusting Medicare Advantage Star Ratings for socioeconomic status and disability. Am J Manag Care. 2018; 24:e285–e291. [PubMed] [Google Scholar]
- [31].Young GJ, Rickles NM, Chou CH, et al. Socioeconomic characteristics of enrollees appear to influence performance scores for Medicare Part D contractors. Health Aff (Millwood). 2014; 33:140–6. doi: 10.1377/hlthaff.2013.0261 [DOI] [PubMed] [Google Scholar]
- [32].National Quality Forum. Risk adjustment for socioeconomic status or other sociodemographic factors. 2014. [Cited 2022 Dec 7]. Available from: http://www.qualityforum.org/Publications/2014/08/Risk_Adjustment_for_Socioeconomic_Status_or_Other_Sociodemographic_Factors.aspx.
- [33].Centers for Medicare & Medicaid Services. Announcement of calendar year (CY) 2017 Medicare Advantage capitation rates and Medicare Advantage and Part D payment policies and final call letter. 2016. [Cited 2022 Dec 7]. Available from: https://www.cms.gov/Medicare/Health-Plans/MedicareAdvtgSpecRateStats/Downloads/Announcement2017.pdf.
- [34].Centers for Medicare & Medicaid Services. Examining the potential effects of socioeconomic factors on Star Ratings. 2015. [Cited 2022 Dec 7]. Available from: https://www.cms.gov/Medicare/Prescription-Drug-Coverage/PrescriptionDrugCovGenIn/Downloads/Research-on-the-Impact-of-Socioeconomic-Status-on-Star-Ratingsv1-09082015.pdf.
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
Medicare databases are United States federal databases sponsored by the Centers for Medicare & Medicaid Services (CMS). These data are available to researchers through the Research Data Assistance Center (ResDAC) at the University of Minnesota, according to a strict protocol for data requests. Users of Medicare databases cannot disclose to, nor share the data with, individuals not listed in the Data Use Agreement. ResDAC can be reached via email at resdac@umn.edu, or by phone at 888-973-7322.