Persistence and Drivers of High-Cost Status Among Dual-Eligible Medicare and Medicaid Beneficiaries

High and increasing health care costs present significant challenges for federal and state governments. One population of substantial interest to policymakers has been dual-eligible beneficiaries, the more than 11 million Americans who are covered by both Medicare and Medicaid. These patients generally are sicker, have lower incomes, and have high levels of comorbidity and disability (1). Consequently, they often have complex needs that make their care particularly expensive. Per capita medical costs for dual-eligible beneficiaries are nearly 4 times higher than those for nondual-eligible Medicare beneficiaries (1). In 2012, although dual-eligible beneficiaries made up only 20% of all Medicare beneficiaries and 15% of Medicaid beneficiaries, they accounted for 34% and 33% of spending for each program, respectively (2). Given how high costs can be among these patients, there has been substantial policy interest in reducing spending for them. Yet, despite this interest and activity, we know surprisingly little about the drivers of spending in this population and even less about the drivers of spending over time. Prior studies suggest that these patients have substantial medical and social burdens that make their care complicated and expensive (1), but we know less about how much of their spending is transient versus persistent. Further, we have little information about high-cost dual-eligible beneficiariesthe most expensive subset in this populationincluding who they are and how their spending patterns differ from those in other populations of dual-eligible beneficiaries. Given the substantial programmatic and policy activity targeting these populations, more granular data on the population are critically needed. In this study, we used national data that comprehensively account for the full costs incurred by these dual-eligible patients to answer the following key questions. First, among those dually eligible for Medicare and Medicaid, what proportion of the population is persistently high-cost across multiple years? Second, what are the characteristics of persistently high-cost patients, and how do they differ from those who are transiently high-cost or nonhigh-cost? Finally, what are the predominant drivers of spending and utilization across persistently high-cost versus transiently high-cost patients, and how much of the spending is related to potentially preventable hospitalizations, which are often a target of intervention by health systems? Methods Data We used MedicareMedicaid Linked Enrollee Analytic Data Source data from 2008 to 2010. The data set contains beneficiary-level Medicare and Medicaid enrollment and claims data for persons enrolled in both programs. For Medicare, the data set contains Part A institutional claims (including inpatient, skilled-nursing facility, home health, and hospice claims), Part B institutional claims (hospital outpatient) and noninstitutional claims (carrier and durable medical equipment), and drug claims. From the Medicaid files, we obtained additional inpatient, outpatient, drug, and long-term care claims. The data set also included information on patient demographic characteristics (age, sex, and race); reasons for enrollment; and the 27 chronic conditions defined by the Chronic Conditions Data Warehouse as well as 25 conditions related to mental health, tobacco use, and physical and mental disability. For each beneficiary, monthly costs were summed for each service category. Use of services was also summed monthly by care setting for each patient. To standardize costs, we divided each cost at the patient level by the wage index used by the Centers for Medicare & Medicaid Services (CMS) to adjust for differences in prices. Wage index data were obtained from the CMS Web site (3). We limited our analysis to patients who were eligible for full Medicare and Medicaid benefits and were alive and dually enrolled for the entire period from 2008 to 2010; we excluded partial-benefit beneficiaries and those who died during the study period. We excluded decedents mainly because we were interested in identifying a population of patients who remained alive and persistently high-cost over time because they may have costs that are potentially more amenable to interventions, such as care management programs, and fewer costs related to potentially high end-of-life spending. As a sensitivity analysis and to evaluate the generalizability of our study, we repeated our analysis with decedents included. To identify potentially preventable hospitalizations, we used the Agency for Healthcare Research and Quality Prevention Quality Indicators software (4), which identifies potentially preventable hospitalizations related to specific conditions, such as heart failure, diabetes, and asthma. These ambulatory caresensitive conditions are believed to be potentially avoidable when appropriate and coordinated outpatient care is delivered to patients. To obtain the ambulatory caresensitive conditions, we used the beneficiary identification number to link the MedicareMedicaid Linked Enrollee Analytic Data Source file with the Inpatient Medicare File from 2008 to 2010. Statistical Analysis We summed annual Medicare and Medicaid payments for all dual-eligible patients for 2008, 2009, and 2010. Next, we categorized beneficiaries as high-cost for each year if their annual expenditures were in the top 10% of total spending among all dual-eligible persons; this meant that each year had a different threshold for the top 10%. We chose this threshold on the basis of a recent National Academy of Medicine report, Effective Care for High-Need Patients, which defined high-need, high-cost populations based on patients who were in the top 10% of spending across the Medicare-only population, the dual-eligible population, and a commercial population (5, 6). We then defined beneficiaries who were high-cost across all 3 years as persistently high-cost and those in the bottom 90% of spending in all 3 years as nonhigh-cost. Patients who were high-cost in 2008 but not in 2009 or 2010 were defined as transiently high-cost. A fourth group, newly high-cost patients, comprised those who were not high-cost in 2008 but became high-cost in either 2009 or 2010. Although newly high-cost patients were not the focus of this study, they were included in several analyses for comparison with persistently high-cost patients. Also, as a sensitivity analysis, we repeated our analysis with decedents included (dual-eligible beneficiaries who died in 2008, 2009, or 2010) and recalculated the top 10% of spending in each year. We then examined what proportion of persistently high-cost patients under the original methods (with decedents excluded) were also persistently high-cost when decedents were included. We did not categorize patients who died in 2008 or 2009 into 1 of the 4 groups because they did not have 3 years of participation, but we did categorize patients who died in 2010. We then performed descriptive comparisons among the groups. First, we compared patient and community demographic characteristics among persistently high-cost, transiently high-cost, and nonhigh-cost patients in all 3 years. Next, we examined differences in spending for inpatient care, outpatient care, postacute care, drugs, and long-term care. Finally, we compared differences in ambulatory caresensitive conditions across the groups. Descriptive analyses were performed using SAS, version 9.4 (SAS Institute). The study was approved by the Harvard T.H. Chan School of Public Health Institutional Review Board's Committee on the Use of Human Subjects. Role of the Funding Source This work was funded by the Peterson Center on Healthcare, which had no role in the design, conduct, or analysis of the study; reporting of the results; or writing of the manuscript. Results Patient Characteristics Our sample consisted of 1928340 dual-eligible beneficiaries. In 2008, 192835 were categorized as high-cost, of whom 132528 (68.7%) remained high-cost in 2009 and 105641 (54.8%) were high-cost in all 3 years (Figure), which suggests that 87194 (45.2%) were transiently high-cost. Figure. Proportions of dual-eligible beneficiaries who remained persistently high-cost versus transiently high-cost and mean yearly spending, 2008 to 2010. In 2008, dual-eligible patients whose spending was in the top 10% (n= 192835) were defined as high-cost. Of these, 54.8% remained persistently high-cost. Of the total sample, 1624257 patients (84.2%) remained nonhigh-cost in all 3 years (Table 1). The remaining 111248 (5.8%) who were originally nonhigh-cost in 2008 became newly high-cost in either 2009 or 2010 (Appendix Figure 1). Table 1. Characteristics of Persistently High-Cost, Transiently High-Cost, and NonHigh-Cost Patients* Appendix Figure 1. Proportions of dual-eligible beneficiaries who remained nonhigh-cost versus newly high-cost and mean yearly spending, 2008 to 2010. In 2008, 1735505 patients were defined as nonhigh-cost. Of these, 93.6% remained nonhigh-cost in the subsequent 2 years. In our sensitivity analysis, we included beneficiaries who died in 2008, 2009, or 2010 and recalculated the proportion who were persistently high-cost. The new sample consisted of 2626785 patients, of whom 330562 died in 2008, 206401 died in 2009, and 161482 died in 2010 (Appendix Table 1). We categorized 262679 patients as high-cost in 2008 (with decedents included); of these, 98360 (37.5%) were persistently high-cost over the subsequent 2 years (Appendix Figure 2). When we compared this with our initial categorization of persistently high-cost without decedents, we found that 92707 (94.3%) of these patients were persistently high-cost under both methods. Appendix Figure 2. Number of patients categorized as high-cost versus nonhigh-cost each year, with decedents included. Appendix Table 1. Sample Size in Each Year for Main Analysis (Decedents Excluded) and Sensitivity Analysis (Decedents Included) We then examined dif