American Journal of Medical Quality

http://ajm.sagepub.com/ Variation in Diabetes Care Quality Among Medicare Advantage Plans : Understanding the Role of Case Mix
Jean M. Abraham, Schelomo Marmor, David Knutson, Jessica Zeglin and Beth Virnig American Journal of Medical Quality 2012 27: 377 originally published online 28 December 2011 DOI: 10.1177/1062860611428529 The online version of this article can be found at: http://ajm.sagepub.com/content/27/5/377

Published by:
http://www.sagepublications.com

On behalf of:

American College of Medical Quality

Additional services and information for American Journal of Medical Quality can be found at: Email Alerts: http://ajm.sagepub.com/cgi/alerts Subscriptions: http://ajm.sagepub.com/subscriptions Reprints: http://www.sagepub.com/journalsReprints.nav Permissions: http://www.sagepub.com/journalsPermissions.nav

>> Version of Record - Sep 10, 2012 OnlineFirst Version of Record - Dec 28, 2011 What is This?

Downloaded from ajm.sagepub.com at UNIV FED DO ESPIRITO SANTO on November 8, 2012

428529
ham et alAmerican Journal of Medical Quality

AJMXXX10.1177/1062860611428529Abra

Article

Variation in Diabetes Care Quality Among Medicare Advantage Plans: Understanding the Role of Case Mix

American Journal of Medical Quality 27(5) 377382 2012 by the American College of Medical Quality Reprints and permission: sagepub.com/journalsPermissions.nav DOI: 10.1177/1062860611428529 http://ajmq.sagepub.com

Jean M. Abraham, PhD,1 Schelomo Marmor, MPH,1 David Knutson, MS,1 Jessica Zeglin, MPH,1 and Beth Virnig, PhD1

Abstract This study investigates whether variation in Medicare Advantage plan performance on comprehensive diabetes care is explained by the case mix of plans. Using data on 513 Medicare Advantage plan-year observations for 2007 and 2008, the authors estimate multivariate regressions for 3 diabetes care quality measures: (1) hemoglobin screening, (2) low-density lipoprotein screening, and (3) retinal eye exam. Plan case mix is measured with the percentage of a plans enrollees who have type 1 diabetes with and without comorbidities and the percentage of a plans enrollees who have type 2 diabetes with and without comorbidities. Plans with a higher percentage of enrollees with type 1 diabetes with comorbidity and plans with a higher percentage of enrollees with type 2 diabetes without comorbidity have lower performance, on average. Finding evidence of a relationship between case mix and Healthcare Effectiveness Data and Information Set performance reinforces the argument for developing standardized risk adjustment or stratification methods in public reporting and pay-for-performance efforts. Keywords diabetes, quality, case mix, Medicare Advantage

Healthcare Effectiveness Data and Information Set (HEDIS) quality scores are used by more than 90% of US health plans to assess performance and to direct quality improvement efforts. Health care quality measurement systems such as HEDIS are being used in public reports of quality information that consumers may access when selecting a health plan. In addition to providing information, these measures are being used to evaluate plans and providers in conjunction with pay-for-performance initiatives.1,2 The recent passage of the Patient Protection and Affordable Care Act further expands the use of these measures. HEDIS measures represent 1 of 4 data sources used as part of the 5-star ratings system for Medicare Advantage plans. Beginning in 2012, Medicare Advantage plans that achieve at least a 4-star rating will earn payment bonuses.3 One criticism of HEDIS measures is that they are not adjusted for underlying differences in enrollee population characteristics. This is problematic to the extent that these differences directly affect performance but are not under the control of the plan or its contracted providers.4,5 Although this criticism may be most applicable to outcome measures such as mortality, it also may be valid for some process-of-care quality measures.6,7

Research documenting the relationship between health plan performance and characteristics of the enrollee population is sparse, given the limited availability of data. To date, studies have examined the relationship between HEDIS quality measures and the socioeconomic characteristics of a plans enrollees but not health-related characteristics. Zaslavsky et al8 used individual-level data on commercially insured members from 10 plans and augmented this information with

University of Minnesota, Minneapolis, MN

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article:The authors received financial support through a grant from Changes in Health Care Financing and Organization, an initiative of the Robert Wood Johnson Foundation. Corresponding Author: Jean M. Abraham, PhD, Division of Health Policy and Management, School of Public Health, University of Minnesota, 420 Delaware Street, SE MMC 729, Minneapolis, MN 55455 Email: abrah042@umn.edu

Downloaded from ajm.sagepub.com at UNIV FED DO ESPIRITO SANTO on November 8, 2012

378 socioeconomic characteristics measured at the zip code level of enrollees (eg, race, ethnicity, education, urban status) to examine the relationship between plan members age, sex, and areas of residence and several HEDIS process-of-care quality measures (ie, checkup after delivery, adolescent immunization, child immu nization, mammogram, Pap smear, prenatal care, retinal exam). Multivariate results found that performance was negatively associated with the percentage of the population in the area who receive public assistance, the percentage black, and the percentage Hispanic. In contrast, performance was positively related to the percentage of the population who were college educated and the percentage who live in an urban area. Although many socioeconomic factors exhibited statistically significant associations with the quality measures, the magnitudes were generally small. Zaslavsky and Epstein9 examined the same question using data on 110 541 commercially insured members in 22 health plans based in California and found results similar to those of Zaslavsky et al.8 They found that adjustment for socioeconomic characteristics changed rates by less than 5 percentage points among most plans, but the effects were considerably larger for a few plans. Trivedi et al10 assessed variations among 151 Medicare Advantage plans in overall quality and racial disparity for 4 HEDIS measures over the 2002 to 2004 time period. Using individual-level data, they examined several quality measures, including having glycosylated hemoglobin (HbA1c) levels of less than 9% for enrollees with diabetes, having low-density lipoprotein cholesterol (LDL) levels of less than 130 mg/dL for enrollees with diabetes after a coronary event, and having blood pressure of less than 140/90 mmHg for enrollees with hypertension. In each model specification, the authors controlled for individual attributes (age, sex, zip codelevel poverty and education, Medicaid enrollment, and year of measurement) and plan characteristics (geographic region, model type, size, percentage black enrollment, plan age, and tax status). The authors found that performance on all outcomes was significantly lower for black enrollees than white enrollees, on average, and that the racial disparity was attributable to both within-plan and between-plan variation. Virnig and colleagues11 used individual-level HEDIS data to examine racial variation in the quality of Medicare managed care. They found significant racial variation in a wide variety of measures, including access to health care,11 mammography use,11 diabetes care,12 and follow-up after hospitalization for mental illness.13 Although prior research has clearly documented a relationship between socioeconomic characteristics of enrollees and HEDIS quality measures, none has investigated the role of enrollees disease type and/or comorbidities. Understanding the importance of these factors in

American Journal of Medical Quality 27(5) explaining variation in HEDIS performance can provide new insights for policy makers and other decision makers because it relates to the debate over risk adjustment or stratification in the measurement and reporting of health plan quality. More broadly, findings from this research have implications for both consumers and health plans. In recent years, there has been a strong push for increased information transparency, including the dissemination of formal sources of provider and plan quality information. One of the most important factors in determining the use of information by consumers is whether the information is perceived as relevant to their circumstances (J.M.A. et al, unpublished data, 2011). To the extent that a health plans performance varies within its enrollee population, reporting an aggregate measure may be misleading. Evidence of a relationship between case mix and HEDIS quality also may affect the behavior of plans, particularly given increased transparency of information and stronger ties of quality to payment. Public reporting or pay-for-performance evaluations that fail to adjust for differences in attributes of enrollees that may influence provision of care would enhance the incentives of these plans to recruit only those enrollees who are easiest served and most compliant.

Methods Study Population and Data

Our primary data source includes HEDIS data reported by Medicare Advantage plans to the Centers for Medicare and Medicaid Services for reporting years 2007 and 2008 (based on 2006 and 2007 experience). The HEDIS data were augmented with plan-level information on the percentage of participants with type 1 and type 2 diabetes with and without comorbidities enrolled in the plan during the year. These data are collected annually as part of the National Committee for Quality Assurances (NCQAs) Relative Resource Use initiative. The sample includes information on 600 plan-year observations, corresponding to 368 unique plans. A total of 68 observations were excluded because of missing values on key variables, and 19 more observations that reported fewer than 100 members with diabetes were removed. The final sample includes 513 plan-year observations.

Measures
Three HEDIS Comprehensive Diabetes Care quality scores were considered to be the outcome measures: (1) the percentage of the eligible diabetes population who had annual HbA1c screening, (2) the percentage

Downloaded from ajm.sagepub.com at UNIV FED DO ESPIRITO SANTO on November 8, 2012

Abraham et al of the eligible diabetes population who had LDL screening, and (3) the percentage of the eligible diabetes population who had a retinal eye exam performed. Eligibility refers to an enrollee being 18 to 75 years of age; having an International Classification of Diseases, Ninth Revision diagnosis code of diabetes and/or being dispensed insulin, oral hypoglycemics, or oral antihyperglycemics during the measurement year or the year prior to that; and having no more than a 1 month gap in coverage. The key explanatory variables capture a plans diabetes enrollee composition during a given year. In the model specification, case mix is defined using 4 indicators: the percentage of enrollees with type 1 diabetes with comorbidity, the percentage with type 1 diabetes without comorbidity, the percentage with type 2 diabetes with comorbidity, and the percentage with type 2 diabetes without comorbidity (reference category). Comorbid conditions are based on NCQAs definition and include asthma, cardiovascular conditions, chronic obstructive pulmonary disease, hypertension, and chronic kidney disease. The model also includes the total number of eligible participants with diabetes enrolled in the plan during the year. Plans with a larger number of enrollees with diabetes are more likely to have invested in quality improvement and/or are more likely to have formal programs in place to assist enrollees with diabetes with managing their condition. Also included in the model is plan type, measured as a binary indicator for whether or not the Medicare Advantage plan is a health maintenance organization (HMO; reference category includes preferred provider organizations [PPOs] and point of service [POS] plans). If HMOs provide better care coordination, we might expect their performance to be higher relative to plans that are not HMOs. Finally, to control for differences by geography and time, indicators corresponding to the 10 US Department of Health and Human Services (HHS) regions were included, consistent with NCQAs standard HEDIS reporting (reference category is region 1) as well as a year indicator dummy for 2008 (reference category is 2007).

379 controls for geographic region and time. Because there are multiple observations on plans, the standard errors were clustered based on the plan identifier. All analyses were conducted using STATA/SE, version 11.0 (StataCorp LLP, College Station, TX).

Results Descriptive Analysis

Table 1 provides summary statistics for the sample. For the 513 plan-year observations, the average rates of HbA1c and LDL screening were high. Performance on retinal eye exams was lower. There was extensive variation in performance across plans for all 3 outcomes. For example, retinal eye exam rates varied from 0% to 91.72%. Of the 4 case-mix categories, the percentage of enrollees who have type 2 diabetes with complications was the most prevalent on average. Plans varied considerably in terms of the composition of their enrolled population with diabetes. For example, the percentage of enrollees who have type 2 diabetes with complications ranged from 1.49% to 87.63% across plans, whereas the percentage of enrollees who have type 1 diabetes without complications ranged from 0% to 79.98%. The Medicare Advantage plans are diverse in terms of size, type, and geography. The minimum number of eligible enrollees with diabetes in a plan during the year was 101, and the maximum was 43 369. A total of 39% were PPOs or POS plans. Plans in the sample represent all 10 HHS geographic regions.

Multivariate Analysis
Table 2 reports parameter estimates and standard errors for the 3 models. Based on a joint F test of all of the slope parameters, the models are statistically significant overall with R2 statistics ranging from 0.24 to 0.29. Across all 3 specifications, there is robust empirical evidence of a relationship between the case mix of a plans diabetes enrollee population and its performance on the quality scores. Plans having a higher percentage of enrollees who have type 2 diabetes with comorbidities are associated with higher performance on average, holding all else constant. The magnitude of the relationship suggests a modest-sized effect. For example, if the percentage of a plans enrollees with type 2 diabetes with comorbidities increased from 5% to 10%, this would be associated with a 1.289 percentage point increase, on average, in its measured HbA1c screening rate. Given the model specification, this implicitly assumes that the percentage of enrollees with type 2 diabetes without

Analysis
The authors begin by providing a descriptive analysis of the sample of Medicare Advantage plans, including summary statistics for the HEDIS outcomes, diabetes enrollee case mix, and plan characteristics. Following this, results from the multivariate analysis are presented. Using ordinary least squares, each quality outcome is regressed on the enrollee case-mix measures reflecting diabetes type and comorbidity prevalence, total number of enrollees with diabetes, and plan type as well as

Downloaded from ajm.sagepub.com at UNIV FED DO ESPIRITO SANTO on November 8, 2012

380
Table 1. Descriptive Statistics Variable Name Percentage of members with diabetes receiving HbA1c screening Percentage of members with diabetes receiving LDL screening Percentage of members with diabetes receiving retinal eye exam Percentage of members with type 1 diabetes with comorbidity Percentage of members with type 1 diabetes without comorbidity Percentage of members with type 2 diabetes with comorbidity Percentage of members with type 2 diabetes without comorbidity Number of eligible members with diabetes in plan during year HMO plan Region 1 (CT, ME, MA, NH, RI,VT) Region 2 (NJ, NY) Region 3 (DE, DC, MD, PA,VA, WV) Region 4 (AL, FL, GA, KY, MS, NC, SC, TN) Region 5 (IL, IN, MI, MN, OH, WI) Region 6 (AR, LA, NM, OK, TX) Region 7 (IA, KS, MO, NE) Region 8 (CO, MT, ND, SD, UT, WY) Region 9 (AZ, CA, HI, NV) Region 10 (AK, ID, OR, WA) Year 2008

American Journal of Medical Quality 27(5)

Mean/Proportion 86.968 84.696 60.644 13.369 2.770 64.631 19.229 2933 0.610 0.031 0.144 0.113 0.138 0.158 0.086 0.049 0.031 0.170 0.080 0.589

Standard Deviation 7.844 7.621 15.149 7.986 5.118 11.542 9.119 5457 0.488 0.174 0.352 0.317 0.346 0.365 0.280 0.216 0.174 0.376 0.271 0.493

Abbreviations: HbA1c, glycosylated hemoglobin; HMO, health maintenance organization; LDL, low-density lipoprotein cholesterol.

Table 2. Multivariate Regression Results for Diabetes Quality Measuresa Percentage Receiving HbA1c Screen Variable Name Percentage Receiving LDL Screen Percentage Receiving Retinal Eye Exam

Parameter Robust Standard Parameter Robust Parameter Robust Standard Estimate Error Estimate Standard Error Estimate Error 0.0876 0.1049 0.0843 0.00005 0.7525 Reference 1.556 1.313 1.304 1.520 1.349 1.258 1.693 1.658 1.418 0.426 6.822 0.0743 0.3502** 0.3506** 0.0002** 2.033** Reference 1.623 4.456** 2.135 4.211** 4.621** 4.999** 7.773** 2.065 2.125 0.441 61.068 ** 0.269 513 0.0789 0.0854 0.0724 0.0005 0.740 Reference 1.423 1.483 1.570 1.635 1.527 1.429 2.064 1.448 1.701 0.454 6.012 0.1132 0.2968* 0.3669** 0.0004** 4.470** Reference 15.300** 12.737** 22.318** 7.465* 19.293** 7.760* 10.697** 18.556** 3.995 0.892 47.895** 0.290 513 0.1237 0.1226 0.1012 0.0001 1.404 Reference 3.111 2.905 3.316 3.063 3.260 3.164 3.856 3.223 3.203 0.832 8.242

Percentage type 1 with comorbidity 0.0021 Percentage type 1 without comorbidity 0.3197** Percentage type 2 with comorbidity 0.2578** Number of eligible members with 0.0002 diabetes in plan during year HMO plan 2.145** Region 1 Reference Region 2 8.582** Region 3 4.755** Region 4 4.897** Region 5 2.765 Region 6 4.254** Region 7 2.339 Region 8 3.180 Region 9 6.546** Region 10 0.454 Year 2008 0.024 Constant 72.119** R2 0.242 Number of observations 513

Abbreviations: HbA1c, glycosylated hemoglobin; LDL, low-density lipoprotein cholesterol; HMO, health maintenance organization. a *P < .05; **P < .01. Region 1: CT, ME, MA, NH, RI,VT; region 2: NJ, NY; region 3: DE, DC, MD, PA,VA, WV; region 4: AL, FL, GA, KY, MS, NC, SC, TN; region 5: IL, IN, MI, MN, OH, WI; region 6: AR, LA, NM, OK, TX; region 7: IA, KS, MO, NE; region 8: CO, MT, ND, SD, UT, WY; region 9: AZ, CA, HI, NV; region 10: AK, ID, OR, WA.

Downloaded from ajm.sagepub.com at UNIV FED DO ESPIRITO SANTO on November 8, 2012

Abraham et al comorbidity (the reference category) is decreasing by 5 percentage points. The magnitude is somewhat larger for the LDL screening rate. Unlike the first 2 measures, there is a negative and large effect of case mix with respect to a plans percentage of enrollees receiving a retinal eye exam. Here, a 5 percentage point increase in enrollees with type 2 diabetes with comorbidities is associated with having an eye exam rate that is 1.835 percentage points lower, on average. In contrast, relative to the percentage of enrollees with type 2 diabetes without comorbidities, a plan having a higher percentage of enrollees with type 1 diabetes without comorbidities is associated with higher retinal eye exam rates. Overall, the case-mix variables explain a significant amount of variation in the performance of plans on the 3 process-of-care measures. In supplemental analyses, the authors reestimated each model excluding the case-mix indicators as covariates. Differences in the R2 statistics between the full and restricted model specifications suggest that including plan-level measures of case mix explains 8%, 14.1%, and 6.4% of variation across plans in HbA1c screening, LDL screening, and retinal eye exam screening rates, respectively. Thus, case mix may be an important predictor of performance. Other variables within the models also exhibit significant relationships to the performance of plans. First, there is a small positive relationship between the total number of eligible participants with diabetes enrolled in a plan during the year and each process-of-care measure. This is consistent with the notion that plans with larger numbers of chronically ill patients may invest more heavily in quality improvement efforts, leading to higher quality scores. Second, there is a large positive effect of HMO plan type on quality. Compared with a PPO or POS plan, an HMO has an LDL screening rate that is approximately 2 percentage points higher, all else being constant. One possible explanation is that the organizational structure of HMOs may allow them to be more effective with respect to care coordination. This includes enrollees having assigned and accountable primary care physicians who can ensure that enrollees with diabetes obtain recommended screenings. Finally, the parameter estimates from the set of geographic region indicators suggest substantial heterogeneity with respect to performance in Medicare Advantage plans. For all 3 outcomes, regions 1 and 10 exhibited the highest performance, whereas the lowest performance for the 3 outcomes varied.

381 demographic characteristics, we find that plans that have a higher percentage of enrollees who have type 2 diabetes without comorbidities or type 1 diabetes with comorbidities are associated with having lower performance on 3 process-of-care measures, including HbA1c screening rates, LDL screening rates, and retinal eye exam rates. Further study is needed to better understand this observed relationship. Several study limitations should be noted. First, diabetes type and presence of comorbidity may be correlated with other enrollee characteristics that are unobserved by the researcher. For example, we know that some demographic factors, such as age, are correlated with diabetes type. Thus, it is not possible to assess whether case mix or other factors drive the underlying association that we seek to measure. Second, even though we found quality to be associated with case mix, there is no assurance that the quality experience of the individual members is similarly associated with case mix. For example, Virnig and colleagues14 used individuallevel HEDIS quality data and found that the percentage of African American enrollees in Medicare managed care organizations was correlated with the quality of care received by white enrollees. Without individual-level data, one cannot assess whether the magnitude and significance of the relationship between case mix and quality persists when evaluated at the individual level. In this article, we have assessed only the association between case-mix characteristics and HEDIS performance for diabetes; patterns between case mix and quality may be different for other diseases. Finally, our evidence is associative rather than causal. Future research is needed to investigate whether case mix exhibits a causal relationship with plan quality.

Conclusion
HEDIS measures are being used by health plans and providers to monitor quality improvement activities, for public reporting, and as part of pay-for-performance initiatives. As part of the Patient Protection and Affordable Care Act, HHS will make quality bonus payments to Medicare Advantage plans that obtain at least 4 stars on the rating systeman initiative that relies in part on the performance of plans on HEDIS measures. Given this, there is renewed interest in ensuring that measures are properly interpretable and in understanding the factors that may explain variation in these quality measures across plans and providers. In this study, we take advantage of newly available data to investigate the extent to which differences in clinical case mix can explain variation in health plan quality for diabetes care. Results from this study find evidence of such an association and suggest that more work is needed to evaluate whether such associations reflect stronger causal relationships.

Discussion
Results from the analysis are consistent with prior research illustrating that variation in plan quality is associated with the case mix of the enrollee population. However, unlike prior research that examines only

Downloaded from ajm.sagepub.com at UNIV FED DO ESPIRITO SANTO on November 8, 2012

382 Acknowledgment
The authors thank Sarah Scholle, DrPH, at the National Committee for Quality Assurance for valuable advice.

American Journal of Medical Quality 27(5)

5. Kassirer JP. The use and abuse of practice profiles. N Engl J Med. 1994;330:634-636. 6. Kuhlthau K, Ferris TG, Iezzoni L. Risk adjustment for pediatric quality indicators. Pediatrics. 2004;113:210-216. 7. Rubin HR, Pronovost P, Diette GB. The advantages and disadvantages of process-based measures of health care quality. Int J Qual Health Care. 2001;13:469-474. 8. Zaslavsky AM, Hochheimer JN, Schneider EC, et al. Impact of sociodemographic case mix on the HEDIS measures of health plan quality. Med Care. 2000;38:981-992. 9. Zaslavsky AM, Epstein A. How patients sociodemographic characteristics affect comparisons of competing health plans in California on HEDIS quality measures. Int J Qual Health Care. 2005;17:67-74. 10. Trivedi AN, Zaslavsky AM, Schneider EC, Ayanian JZ. Relationship between quality of care and racial disparities in Medicare health plans. JAMA. 2006;296:1998-2004. 11. Virnig BA, Lurie N, Huang Z, Musgrave D, McBean AM, Dowd B. Racial variation in quality of health care provided by Medicare+Choice health plans. Health Aff (Millwood). 2002;21:224-230. 12. McBean AM, Huang Z, Virnig BA, Lurie N, Musgrave D. Racial variation in the control of diabetes among elderly Medicare managed care beneficiaries. Diabetes Care. 2003;26:3250-3256. 13. Virnig BA, Huang Z, Lurie N, Musgrave D, McBean AM, Dowd B. Does Medicare managed care provide equal treatment for mental illness across races? Arch Gen Psychiatry. 2004;61:201-205. 14. Virnig BA, Scholle SH, Chou AF, Shih S. Efforts to reduce racial disparities in Medicare managed care must consider the effects of geography. Am J Manag Care. 2007;13:51-56.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The authors received financial support through a grant from Changes in Health Care Financing and Organization, an initiative of the Robert Wood Johnson Foundation.

References
1. Kuhmerker K, Hartman T. Pay-for-performance in state Medicaid programs: a survey of state Medicaid directors and programs. http://www.providersedge.com/ehdocs/ehr_ articles/Pay-for-Performance_in_State_Medicaid_Programs.pdf. Accessed March 23, 2011. 2. Rosenthal MB, Landon BE, Normand ST, Frank RG, Epstein AM. Pay for performance in commercial HMOs. N Engl J Med. 2006;355:1895-1902. 3. Jacobson G, Damico A, Neuman T, Huang J. Quality ratings of Medicare Advantage plans: key changes in the health reform law and 2010 enrollment data. http://www.kff.org/ medicare/upload/8097.pdf. Accessed March 24, 2011. 4. Jost TS. Health system reform: forward or backward with quality oversight? JAMA. 1994;271:1508-1511.

Downloaded from ajm.sagepub.com at UNIV FED DO ESPIRITO SANTO on November 8, 2012