Medicare beneficiaries with diabetes who are at the lowest levels of healthcare consumption often become some of the highest level consumers in subsequent years.
To examine the annual cost profiles of Medicare beneficiaries with diabetes to identify patterns in their consumption of benefits.
Retrospective expenditure data were collected from Medicare records. Beneficiaries with diabetes were grouped into 5 consumption clusters ranging from “crisis consumers” at the high end to “low consumers” at the low end.
The percentages of beneficiaries and expenditures for the consumption clusters remained generally constant from year to year. As expected, most of Medicare’s budget each year was spent on crisis, heavy, and moderate consumers. However, a notable proportion of low and light consumers from one year go on to become crisis and heavy consumers in subsequent years. A review of total 2001 through 2006 inpatient costs for the year 2000 clusters revealed that 47% of these costs were for year 2000 low and light consumers and only 27% were for year 2000 crisis and heavy consumers.
This analysis revealed previously unrecognized trends, whereby a notable proportion of low and light consumers during one year went on to become crisis and heavy consumers in subsequent years, representing a large proportion of inpatient costs. These findings have important implications for disease management programs, which typically focus intervention efforts exclusively on crisis and heavy consumers.
Am J Manag Care. 2013;19(7):541-548This study analyzed annual cost profiles of Medicare beneficiaries with diabetes to identify patterns in their consumption of benefits.
Previous studies have demonstrated that a relatively small proportion of Medicare beneficiaries are the principal consumers of Medicare benefits.1-6 By comparing healthcare expenditures over a number of years, Berk and Monheit2 noted that there is a large amount of stability in the spending distribution over time. Because of the stable, skewed distribution of healthcare expenditures, the prevailing approach to reducing healthcare spending has been to focus on those who are receiving large amounts of care.
However, it has recently become clear that disease management programs that focus on those receiving the largest amounts of care have not been successful. A recent report from the Congressional Budget Office describes an evaluation of Medicare’s demonstration projects on disease management.7 Six major disease management and care coordination demonstrations were evaluated by independent researchers. The results showed that most programs did not reduce Medicare spending, despite the fact that the programs specifically targeted beneficiaries who were high users of healthcare services.7
Our study extends the previous work by reviewing consumption and hospitalization patterns for fee-for-service Medicare beneficiaries living with diabetes. Consistent with previous findings, the proportions of patients in the various spending clusters were stable over time. However, when the behavior of individual beneficiaries was studied further it became clear that the populations in the spending clusters were dynamic, reconstituted each year while retaining the same proportional dimensions, as beneficiaries migrated from 1 cluster to another within definable parameters. These findings show that there is a previously unreported trend for beneficiaries in the lowest-spending clusters to transition to the highest-spending clusters in subsequent years. We discuss the potential implications these findings for future approaches to reducing healthcare spending.
RESEARCH DESIGN AND METHODS
Retrospective expenditure data were collected from Medicare records. Medicare patients with diabetes were identified with an algorithm that required at least 1 inpatient, skilled nursing facility, or home health agency claim, or 2 Health Options Program or carrier (defined as Medicare administrative contractors or carriers that oversee the administration of both Medicare part A and part B policies) claims with the following diagnosis codes (any diagnosis on the claim) during the 2-year period: 249.00, 249.01, 249.10, 249.11, 249.20, 249.21, 249.30, 249.31, 249.40, 249.41, 249.50, 249.51, 249.60, 249.61, 249.70, 249.71, 249.80, 249.81, 249.90, 249.91, 250.00, 250.01, 250.02, 250.03, 250.10, 250.11, 250.12, 250.13, 250.20, 250.21, 250.22, 250.23, 250.30, 250.31, 250.32, 250.33, 250.40, 250.41, 250.42, 250.43, 250.50, 250.51, 250.52, 250.53, 250.60, 250.61, 250.62, 250.63, 250.70, 250.71, 250.72, 250.73, 250.80, 250.81, 250.82, 250.83, 250.90, 250.91, 250.92, 250.93, 357.2, 362.01, 362.02, and 366.41.
An annual cost profile for each Medicare fee-for-service beneficiary with diabetes for the years 2000 through 2006 was constructed using a unique beneficiary identifier to link the Beneficiary Annual Summary Files to Medicare’s Chronic Condition Data Warehouse flags, which identified beneficiaries who received Medicare reimbursements for diabetes.8 Total annual Medicare expenditures for each beneficiary with diabetes were calculated as the sum of all reimbursements made for inpatient and outpatient care, skilled-nursing facilities, carriers, durable medical goods, and home health and hospice care during a calendar year. In 2000, there were 106,995 (2.1%) diabetes beneficiaries whose primary payer for their medical expenses was not Medicare. Medicare total reimbursements therefore did not reflect their real insured healthcare expenditures. We removed such beneficiaries from study for the years 2000 through 2006. In 2000, there were 1376 (0.03%) diabetes beneficiaries with negative Medicare reimbursement values. Among these beneficiaries, 141 were removed because Medicare was not the primary payer of their medical expenses. For each year, the study included beneficiaries enrolled in Medicare part A and part B for at least 10 months and excluded all beneficiaries enrolled in Medicare Advantage because of the lack of complete reimbursement data. Beneficiaries were only removed from the study when they had at least 1 month of managed care for the year. The date of the death of a beneficiary was determined by the date given in the summary file.
All analyses were conducted with SAS software, version 9.1 (SAS Institute Inc, Cary, North Carolina). On the basis of their annual cost profiles, beneficiaries with diabetes were grouped into 5 consumption clusters. Simple percentages were used to describe patterns of Medicare expenditures for the clusters: (1) crisis consumers (beneficiaries accounting for the 99th percentile [top 1%] of aggregate Medicare payments); (2) heavy consumers (90th through 98th percentiles); (3) moderate consumers (75th through 89th percentiles); (4) light consumers (50th through 74th percentiles); and (5) low consumers (1st through 49th percentiles).
An analysis was performed to determine whether the annual repopulation of a cluster followed any discernible pattern. Percentages were calculated to ascertain what proportions of the beneficiaries in a cluster migrated from one of the prior year’s consumption clusters. For example, what percentages of crisis consumers in 2001 were crisis, heavy, moderate, light, or low consumers in 2000? These migration studies were conducted for each of the clusters for the years 2001 through 2006. Once these calculations were completed, a comparison was made to determine whether the migration patterns among clusters were stable, with similar proportions from one year to the next.
The influence that expenditures for inpatient care had on migration patterns was examined by calculating reimbursements for hospitalization for the 5 consumption clusters from 2000 through 2006. Percentages were derived by dividing total expenditures for inpatient care by the total of all expenditures in a given year.
The annual risk for hospitalization for each of the 5 clusters in a given year was obtained by following each cluster from each year as a distinct cohort. For example, crisis consumers in the year 2000 were analyzed for their risk for inpatient stays in the succeeding years (2001-2006). Cluster members who did not remain fee-for-service beneficiaries from 2000 through 2006 were excluded from the analysis. Hospitalization in a given year was defined as a yes/no variable based on whether or not a beneficiary in a cluster had a record of any inpatient stay during a calendar year; a logistic regression analysis was conducted to determine an annual aggregated risk for hospitalization for each cluster based on its members’ prior histories of hospitalization.
To distinguish the percentage of total inpatient costs for the years 2001 through 2006 attributable to a particular year 2000 consumption cluster, the calculation included the sum of all inpatient costs for the years 2001 through 2006 for a cluster (the numerator) divided by all hospitalization costs for the same period for all year 2000 clusters (the denominator).
shows the annual expenditures and populations of each consumer cluster. As expected, a small proportion of the population (ie, the crisis and heavy consumers) represented the majority of the expenditures (). Analysis of the clusters over time revealed that they were stable in the sense that the percentages of benefi ciaries and expenditures that differentiated each cluster remained generally constant from one year to the next (). For example, crisis consumers ranged between 2% and 3% of all beneficiaries, and 23% to 25% of total reimbursements were for their care.
Of all the clusters, low consumers were the most numerous. Each year they accounted for approximately 32% of all fee-for-service beneficiaries living with diabetes. Total reimbursements for this cluster were approximately 2% of all dollars spent on diabetes in the program. Together, low and light consumers represented 61% of Medicare beneficiaries living with diabetes, but only 9% of the program’s dollars went to their care each year. The mean per capita expenditure for low consumers was $494 in 2000, increasing to $815 in 2006; the corresponding expenditures for crisis consumers were $95,847 in 2000 and $126,789 in 2006. Of total reimbursements, 91% were spent on crisis, heavy, and moderate consumers, who represented 39% of beneficiaries.
The populations within the clusters were dynamic, reconstituted each year as beneficiaries migrated from one cluster to another. Migration was not unidirectional; some benefi ciaries moved to higher-cost clusters and some moved to lowercost clusters. The percentages of beneficiaries moving among clusters had stable patterns. summarizes annual migration into the crisis-consumer and low-consumer clusters. Discernible patterns were evident. Each year, for example, approximately 20% of crisis consumers remained in that cluster, and about 8% migrated to become low consumers. Routinely, about 39% of crisis consumers had been heavy consumers in the prior year. About 60% of low consumers remained in the cluster from one year to the next. Fewer than 1% of low consumers had been crisis consumers in the previous year.
Each consumption cluster exhibited a unique pattern of risk for future hospitalizations. Logistic regression analysis was used to estimate annual hospitalization risk of the year 2000 consumption clusters for 2001 through 2006 (). The analysis was predicated on members’ having been hospitalized each year, with the exception of low and light consumers, who for the most part were not hospitalized in 2000. A year 2000 low consumer had a 15% chance of being hospitalized in 2001, but with that hospitalization the risk for inpatient care in 2002 rose to 30%. Two consecutive years of hospitalization brought the risk to 43% in 2003. By 2006 the risk had grown to 69%. year 2000 crisis consumers had the highest risk for inpatient care, beginning with 58% in 2001 and culminating at 89% in 2006. For all beneficiaries, consecutive years of hospital care raised future risks. If there was an intervening year or years in which there was no hospitalization, new risk patterns that were unique to the clusters were found. Similar longitudinal analyses were conducted for the clusters starting with other years studied; there was no significant change in risk.
Most of Medicare’s budget each year was spent on crisis, heavy, and moderate consumers. Nevertheless, an important trend was seen when analyzing total hospitalization costs between 2001 and 2006 for the year 2000 clusters. This analysis revealed that 47% of all inpatient costs from 2001 through 2006 were for year 2000 low and light consumers and only 27% were for year 2000 crisis and heavy consumers (Table 4, Figure 3). Members of clusters in 2001 and 2002 exhibited similar patterns in succeeding years.
This study aggregated Medicare beneficiaries with diabetes into consumption clusters and found that they consumed future benefits within measurable parameters. These clusters were stable in the sense that the percentages of beneficiaries and expenditures that differentiated each cluster remained generally constant from one year to the next, supporting previous observations that a small number of beneficiaries consume more than 70% of the program’s diabetes budget. These findings are also consistent with more general healthcare spending analyses, which show a large amount of stability in the spending distribution over time.2
Interestingly, clusters’ populations were dynamic, reconstituted each year while retaining the same proportional dimensions as beneficiaries migrated from one cluster to another within definable parameters. Although the annual migration patterns were proportionally stable, they were not unidirectional: beneficiaries migrated to less expensive as well as to more expensive clusters. Each cluster was associated with a specific risk pattern for future hospitalizations.
These findings have potentially important implications for future approaches to reducing healthcare spending. Recent findings show that disease management and care coordination demonstration programs that focus interventions on the highest-consuming clusters did not reduce Medicare spending. The findings in the current study indicate that low consumers represent a significant proportion of future high-expenditure patients. Therefore, research should be conducted to identify the characteristics of the low-consuming beneficiaries who subsequently become high-consuming beneficiaries. Healthcare costs may be significantly reduced by focusing intervention efforts on these high-risk, low-consuming patients.
Although the immediate purpose of this study was to document the stability of consumption patterns among all fee-for-service beneficiaries, it should be noted there were slight differences in diabetes consumption group patterns between disabled patients and those without a disability. The differences were mainly found in the percentages who were crisis, light, and low consumers. For example, 4.4% of all diabetes beneficiaries with a disability were crisis consumers compared with 2.6% of diabetes beneficiaries who had no disability. The majority of the crisis consumers (82%), however, had diabetes without any disabilities. The disabled/nondisabled variation suggests that further research might find subgroups within the main 5 consumption groups that make some contribution to the larger migration patterns characterized in this study.
One potential limitation of this study involves data acquisition for Medicare-enrolled patients. Data for 100% of Medicare-enrolled beneficiaries are available from the Chronic Condition Data Warehouse. The Beneficiary Summary File is created annually and contains demographic and enrollment data for all beneficiaries who are alive and enrolled in Medicare for any part of the year. This file is available in the current layout for 1999 forward. Therefore, the Chronic Condition Data Warehouse data used in this study should include all Medicare-enrolled beneficiaries from 2000 through 2006.9 A second limitation of this study involves the algorithm used to identify Medicare beneficiaries with diabetes. Although the algorithm is adequately sensitive, highly specific, and reliable,there is still some probability that beneficiaries with diabetes were not identified by the algorithm (type II error).10 This error could influence the results, but it is unlikely to alter the study conclusions.
It is possible that a proportion of the low consumers are not receiving adequate medical care. A patient with diabetes should be managed according to clinical guidelines, which help delineate appropriate consumption of essential medical products and services according to a specific timetable to achieve glucose control and to ensure the necessary screenings to detect the onset of complications. A cost structure is associated with these products and services. At a minimum, a well-managed patient with diabetes should have an annual expenditure pattern (cost profile) reflecting that cost structure. A diabetes patient whose cost profile falls significantly below that minimum may be underconsuming, according to established recommendations for optimal care. Future research will need to determine whether the cost profile of low consumers meets those minimum requirements or whether underconsumption may be raising their risk for future inpatient care. Alternatively, it may be that a low consumer is being managed according to clinical guidelines and that the 15% yearly risk for hospitalization and upward migration simply expresses the progressive state of the disease and the somewhat limited long-term impact of current treatment strategies and technology, such as lifestyle modification programs, pharmacologic interventions, and devices.
Additional research is needed to identify the factors that influence the migration of low consumers into more expensive clusters, where the potential for inpatient care in succeeding years increases dramatically. If this upward migration of low consumers can be retarded (by improved technology, better care management, or both), short-term and medium-term cost reductions in the Medicare program might be achieved.Author Affiliations: From Joslin Diabetes Center, Harvard Medical School (AEC), Boston, MA; University of Texas Southwestern Medical School (JD), Dallas, TX; George Washington University School of Public Health and Health Services (AE), Washington, DC; Healing Our Village (JG), Lanham, MD; Commonwealth Family Practice Group (KL), Jacksonville, FL; Howard University (GLN-B), Washington, DC; George Washington University School of Medicine (FZ), Washington, DC; National Minority Quality Forum (GAP), Washington, DC.
Funding Source: None.
Author Disclosures: The authors (AEC, JD, AE, JG, KL, GLN-B, FZ, GAP) report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.
Authorship Information: Concept and design (AEC, AE, JG, KL, GLN-B, FZ, GAP); acquisition of data (FZ, GAP); analysis and nterpretation of data (AEC, JD, AE, JG, KL, GLN-B, GAP); drafting of the manuscript (AEC, AE, JG, FZ); critical revision of the manuscript for important intellectual content (AEC,JD, AE, JG, KL, FZ, GLN-B); statistical analysis (AE); obtaining funding (GAP); and supervision (GAP).
Address correspondence to: Gary A. Puckrein, PhD, National Minority Quality Forum, 1200 New Hampshire Ave, NW, Ste 575, Washington, DC 20036. E-mail: email@example.com. Berk ML, Monheit AC. The concentration of health expenditures: an update. Health Aff (Millwood). 1992;11(4):145-149.
2. Berk ML, Monheit AC. The concentration of health care expenditures, revisited. Health Aff (Millwood). 2001;20(2):9-18.
3. Berk ML, Monheit AC, Hagan MM. How the U.S. spent its health care dollar: 1929-1980. Health Aff (Millwood). 1988;7(4):46-60.
4. Monheit AC. Persistence in health expenditures in the short run:prevalence and consequences. Med Care. 2003;41(7)(suppl):III53-III64.
5. Yu WW, Ezzati-Rice TM. Concentration of Health Care Expenditures in the U.S. Civilian Noninstitutionalized Population. Statistical Brief 81. Rockville, MD: Agency for Healthcare Research and Quality; 2005.
6. Cohen SB, Yu W. The Persistence in the Level of Health Expenditures Over Time: Estimates for the U.S. Population, 2002-2003. Rockville, MD: Agency for Healthcare Research and Quality; 2006.
7. Lessons From Medicare’s Demonstration Projects on Disease Management, Care Coordination, and Value-Based Payment. Washington,DC: Congressional Budget Office; 2012.
8. Chronic Condition Data Warehouse User Guide: Version 1.8. Arlington, VA: Buccaneer, A Vangent Company; 2011.
9. Schneider K, Roozeboom M, Brenton M; for Buccaneer, A Vangent Company. Chronic Condition Data Warehouse. Getting Started withCMS Medicare Administrative Research Files —A Technical Guidance Paper. Version 1.0 http://www.ccwdata.org/cs/groups/public/documents/document/techguidanceadminresearchfiles.pdf. Published December 2011. Accessed March 28, 2012.
10. Hebert PL, Geiss LS, Tierney EF, Engelgau MM, Yawn BP, McBean AM. Identifying persons with diabetes using Medicare claims data. AmJ Med Qual. 1999;14(6):270-277.