- Email: [email protected]
Capitation for cardiologists: Accepting risk for coronary artery disease under managed care
Ca itation for Cardiologists: Accepting Ris e for Coronary Artery Disease Under Managed Care Robert L. McNamara, MD, MHS, Neil R. Powe, MD, MPH, MBA, Thomas Shaffer, BcS, David Thiemann, MD, Wendy Weller, MHS, and Gerard Anderson, PhD Patients with chronic disease may be excluded from capitated managed care plans due to higher than average expected costs. In an attempt to remedy this inequity, one type of risk adjustment techni ue pro ses to T I Inesses, set se rate capitation rates for certain cll romc in& r ing coronary artery disease (CAD). Cardiologists, who increasingly are requested to accept capitation, will benefit from understanding the impact of using clinical factors as opposed to using demographic factors to set capitation rates. Using a 5% national random sample of the 1992 Medicare population, we determined mean annual expenditures and variation in expenditures of individuals with CAD. We corn red the use of 2 demographic factors currently used ror capitation rate adiustment (age and gender) with 2 factors not currently used-3-digit International Classification of Disease (ICD-9) code (a measure for severity) and Charlson index (a measure for comorbidity). Mean annual expenditures for individuals with CAD were more than double mean
annual expenditures for the general Medicare pop& tion ($6,944 vs $3,247). Among individuals with CAD, mean ex nditures of subgroups defined by both age and gen cf= er ranged from $6,205 to $7,724. In comparison, stratifying by measures of severity and comorbidity identified subgroups with lower and higher mean ex nditures, producing a range of $1,702 to $19,959. Sur stantial variation of expenditures for individuals within subgroups defined by severity and comorbidity remained, with few patients having substantially higher expenditures than the rest. When capitation rates are set with the use of demographic factors alone, patients may be subiected to risk selection and physicians to financial loss. Using clinical measures may decrease the incentive for patient risk selection, but substantial financial risk to physicians would remain, because of a relatively few patients with high expenditures (or costs). 01998 by Excerpta Medica, Inc. (Am J Cardiol 1998;82:1178-1182)
apitation is becoming a more common payment method not only to managed care organizations C but also to physician groups.lJ One-third of United
payments, more comprehensive risk adjustment systems classify individuals into an expected cost category based on demographic and clinical factors.lO-l* We examined 1992 Medicare expenditures for individuals with coronary artery disease (CAD) to compare mean expenditures and variation in expenditures of subgroups defined by demographic factors with those defined by severity and comorbidity, which are particularly relevant to small physician groups who have limited ability to spread financial risk.
States (US) physicians had at least 1 capitated contract in 1995.3 Although capitation at the physician level has historically been reserved for primary care physicians, certain specialists, including cardiologists, are now entering into capitated arrangements as well.’ In theory, capitation provides a financial incentive for physicians to deliver care more efficientlye Unfortunately, capitation provides a disincentive to select individuals with chronic illnesses who have high expected costs.536Adjusting capitated rates to more accurately reflect the expected cost of care has been proposed to protect sicker patients from risk selection, as well as to protect physicians with capitation contracts who care for them from excessive financial 10ss.~,~,s In 1 method of risk adjustment, physicians receive a separate, usually higher, capitation payment for each individual with an identified chronic illness that is expensive to treat.” To further adjust capitation From the Departmen% of Epidemiology, Departmen. of Medicine, and Hea,th PO icy and Mona ement, Tie Johns Hopkins Medical Institutions, Bal-imore, Marylan 3 This study was supported by the Commonwealth Fund Grant #96169, New York, ‘\lew YOK. Manuscript received February 9, 1998; revised manuscripr rece’ved and accepted JLne 19, 1998. Address for rep-ints: Robert 1. McNamara, MD, MHS, School of Hygiene and Public Health, Johns Yopkins Jn’vexity, 65 %orth Wolfe Sheet, Room 6009, Baltimore, Mayland 21205. E-mail: [email protected].
1178
01998 by Excerpta Medica, All rights reserved.
Inc.
METHODS Patient selection: Data were obtained from standard research files from the Health Care Financing Administration, consisting of a 5% nationally random sample of Medicare beneficiaries. The enrollment file contains demographic information on each Medicare beneficiary, and the claims file contains expenditures and utilization at the individual beneficiary level for all Medicare covered services. Medicare Part A claims included hospital inpatient and hospital outpatient claims. Medicare Part B included physician inpatient and outpatient services as well as supplier services. International Classification of Disease (ICD-9) codes for services rendered were included with the expenditure data. We combined the enrollment and claims files to develop individual records for each Medicare beneficiary in our sample. We excluded Medicare beneficiaries <65 years of age, who typically have permanent disability or endstage renal disease and may confound a resource anal0002.9149/98/S 19.00 PII SOOO2-9149(98)00602-X
CAD (n = 56,820), without any Medicare Part B claims involving CAD (n = 4,278), and with claims of zero dollars (n = 3,042) were excluded. Thus, the General Medicare Coronary Artery Disease final CAD- data set included claims from 250,514 % % beneficiaries. Age b) Persons with CAD were classified further into 1 of 65-69 28.0 21.0 5 specific diagnoses based upon the 5 3-digit codes 25.2 23.7 70-74 (410 to 414) that were used because the fourth and 75-79 20.5 22.5 fifth digits are often missing. If a person had two 410 14.1 17.0 80-84 12.3 15.8 85+ codes, regardless of other CAD codes, they were Gender classified into the 410 subgroup. If a person did not 61.8 54.5 Women have two 410 codes but did have two 411 codes, they 32.2 45.5 Men were classified into the 411 subgroup. The other 3 Charlson Index subgroups were similarly defined in this hierarchical 44.4 24.7 0 1 30.0 28.8 fashion. 2 14.7 21.7 Comorbidity definition: The Charlson index is a 3+ 10.9 24.8 method of classifying individuals by comorbid conditions that alter the risk of mortality.rs Each of 17 comorbid conditions is assigned a value from 1 to 6. For $6,W instance, chronic obstructive pulmo$7,000 nary disease is assigned a weight of 1, and metastatic solid tumor is as$6,000 signed a weight of 6. The Charlson index for an individual is the sum of Mean $5,000 these values for the comorbidities of Expenditures $4,000 that individual. Deyo and col1992 leaguesI6 adapted this index, which $3,000 i was based on review of medical $2,000 records, for use with administrative databases.We applied this method to $1,000 the Medicare files to provide a Charlson index score for each beneficiary. $0 : For simplicity, and to achieve comGeneral Medicare Individuals with parable numbers in each group, we Population Coronary Artery Disease chose to combine all the Charlson indexes of 23 into 1 group. FIGURE 1. Mean expenditures for he general Medicare population (n = 1,221,615) Statistical analysis: The basic anaversus individuals with CAD (n = 250,514). lytic strategy was to compare mean expenditures and variation in expenysis. Beneficiaries living outside the US, those who ditures between different subgroups of Medicare bendid not have Part A and Part B coverage, and those eficiaries. Because statistically significant differences who were enrolled in a managed care plan were not may not be clinically significant when numbers are included due to incomplete utilization information. large, the comparisons were illustrated graphically. Thus, our final data set included 1,221,615 Medicare beneficiaries >65 years old living in the US with RESULTS Individuals with CAD who met our criteria (n = fee-for-service coverage in 1992. The claims files were further refined by excluding 250,514) were older, more likely men, and had more expenditure data from skilled nursing homes, hospice comorbidity than the general Medicare population care, and home health care. Although some managed (n = 1,221,615) (Table I). Per capita annual expendicare organizations cover these services, they are often tures by the Medicare program for individuals with paid by Medicare on a fee-for-service basis. In addi- CAD were more than double those of the general tion, how plans define and structure these benefits Medicare population ($6,944 vs $3,247) (Figure 1). varies across plans. ‘3 Diagnostic and procedural codes Such a difference may provide a financial disincentive from all other covered services were reviewed. to enroll individuals with CAD unless payment rates Coronary artery disease diagnosis definition: This reflect the higher expected cost. Therefore, establishment of a separatecapitation rate for individuals with was identified by the ICD-9 Clinical Modification (ICD-9-CM) codes 410 to 414. Because CAD fre- CAD is an attractive proposal to minimize this disinquently is misclassified in claims databasesi benefi- centive. The shape of the curve illustrating the distribution ciaries with CAD were defined as having a CAD code at least twice, with 2 1 claim from Medicare Part B. of expenditures of individuals with CAD was similar Beneficiaries without 2 separate claims involving to that of the general Medicare population (Figure 2). TABLE I Comparison Population
Between the General and Individuals With Coronary
Medicare Artery Disease
CORONARY ARTERY DISEASE/CAPITATION
FOR CORONARY ARTERY DISEASE
1179
factors that predict costs better, the incentive to select individuals with low expected costs and to avoid indi$50,000 -viduals with high expected costs would remain. $40,000 -In addition to demographic inforExpenditures mation, some clinical information is $30,000 -1992 available in Medicare claims files. Clinical information from the previ$20,000 -ous year or years could be used prospectively to set the subsequent $10,000 -year’s capitation rates. We used diagnostic codes available on Medicare $0 claims files to stratify individuals 100% 80% 40% 80% 0% 20% with CAD. As a measure of severity, Percentile of Population persons were stratified by 3-digit ICD-9 diagnostic codes for CAD. For FIGURE 2. Distribution of expenditures by percentile for the general Medicare popuinstance, acute or recent myocardial lation (n = 1,221,615) and for individuals with CAD (n = 250,514). infarction is ICD-9 code 410. In addition, we stratified the beneficiaries based on the Charlson index, a wellvalidated indicator of comorbid$8,000 ity.15J6 Mean expenditures of persons $7,000 differed substantially when stratified by these measurements of severity $6,000 and comorbidity (Figure 4). Persons with acute myocardial infarction and a Exp~~~rures $5’ooo Charlson index of ~3 (high comor1992 $4,000 bidity) had the highest mean expenditures ($19,959) and those with un$3,000 specified atherosclerosis and no co$2,000 morbidities had the lowest ($1,702). Variation in expenditures: Adjust$1,000 ing the capitation rate only to reflect $0 mean expenditures within subgroups, however, may not protect the provider 75-79 60-64 as+ 65-69 70-74 from variation in cost. We examined how the measurementsof severity and Age comorbidity affected variation in exFIGURE 3. Mean expenditures of subgroups of individuals with CAD stratified by age penditures by illustrating mean expenand gender. For women (white bars): 65 to 69 years old, n = 23,473; 70 to 74 ditures by percentile for each meayears old, n = 28,874; 75 to 79 years old, n = 30,314, 80 to 84 years old, n = sure. Stratifying these measurements 25,860; and 285 years old, n = 28,117. For men /black bars): 65 to 69 years old, did little to change the shape of the n = 29,036; 70 to 74 years old, n = 30,552; 75 to 79 years old, n = 26,026; 80 to 84 years old, n = 16,707; and ~85 years old, n = 11,555. distribution of expenditures (Figure 5 compared with Figure 2). A few individuals continued to have substanAlthough the expenditures of most people were below tially higher expenditures in each subgroup. Thus, the mean for each group, the expenditures for a few rates that account for severity and comorbidity in this were considerably higher. Thus, establishing a sepa- way would not protect the physician who accepts rate capitation rate for individuals with CAD would capitation from substantial financial loss due to a few not eliminate the financial risk to providers because of patients with high cost expenses. a relatively few patient with high cost expenses. Factors affecting cost: Because age and gender are DISCUSSION Managed care organizations increasingly are using currently used by Medicare to set rates for beneficiaries enrolling in managed care organizations and by capitation as a means to compensatephysicians. Large some managed care organizations to set capitation administrative databases, such as Medicare expendirates to physicians, we examined mean expenditures ture files, are likely to be a source of data in setting of individuals with CAD stratified by these factors. payment rates. Currently, Medicare pays managed However, this method did not substantially differen- care organizations a per capita rate of 95% of the tiate mean expenditures among groups. Mean expen- expected costs for the average fee-for-service benefiditures ranged from $6,205 to only $7,724 (Figure 3). ciaryr” with adjustments for age, gender, welfare staIf payers or capitated providers know of additional tus, institutional status, and eligibility status of bene$80,000 -
ii80
THE AMERICAN JOURNAL OF CARDIOLOGY@
VOL. 82
NOVEMBER
15, 1998
FIGURE4. Mean expenditures of subgroups of individuals with CAD stratified by measures of severity and comorbidity. 4 10 = acute or recent myocardial infarction (n = 30,461); 411 = unstable angina (n = 26,587); 412 = prior myocardiaf infarction (n = 8,004); 413 = stable angina (n = 46,289); and 414 = unspecified atherosclerosis (n = 139,173). Charlson index of 0 (n = 71,748), Charlson index of 1 (n = 74,797), Charlson index of 2 (n = 52,31 l), and Char&on index ~3 (n = 50,658).
$80,000
T 410
$70,000
411
$60,000 Expenditures 1992
J-Digit ICD-9 412 Code 413 414
$50,000 $40,000 $30,000 $20,000 $10,000 $0
0%
20%
40%
60%
80%
100%
Percentile of Population A $100,000
T
I 3+
$80,000 ! II
0%
20%
40%
60%
80%
2 Charlson Index
100%
Percentile of Population B FIGURE 5. Distribution of expenditures by percentile for individuals with CAD stratified by /A) severity and (Bj comorbidity. See Figure 4 for numbers in each group.
ficiaries.r7 Capitation rates based on expected costs of the average Medicare beneficiary discourages managed care organizations and physicians from caring for people with chronic illnesses such as CAD due to the higher expected costs of this population. Setting different capitation rates based upon an individual’s principal diagnoses is 1 type of risk-adjustment method. Age and gender, 2 of the factors currently used to set capitation rates for Medicare beneficiaries enrolling in managed care organizations, did not identify subgroups of individuals with CAD with very high or very low mean expenditures. However, the other factors available in large administrative databases that we examined-measurements of severity and comorbidity-did identify subgroups with high and low mean expenditures. Thus, the use of demographic factors would do little to decreasethe financial incentive to avoid individuals with high expected cost, but the use of measurements of severity and comorbidity likely would decreasethis incentive. Variation in cost is particularly relevant to small cardiology groups with limited ability to spread financial risk. With 10% of the general Medicare population accounting for 70% of the total expendituresI a few patients with high costs could lead to financial loss for providers accepting capitation. Unfortunately, our results indicate that disregarding CAD does not limit variation in expenditures either as a whole group or stratified by measurements of severity or comorbidity (Figures 2 and 5). Thus, the use of these measurements of severity and comorbidity would not protect physicians who accept capitation from substantial financial loss due to a few patients with high cost expenses. Reinsurance: Alternative risk-adjustment methods other than diagnostic carveouts are available. Reinsurance, or stop-loss insurance, is one of the most common and covers unexpected losses from extremely high cost expenses.Basically, the individual or group accepting a capitated payment is responsible for annual costs only up to a predefined level. Above that level, the reinsurer assumestotal or partial financial re-
CORONARY ARTERY DISEASE/CAPITATION
FOR CORONARY ARTERY DISEASE
118 1
sponsibility. For instance, a group of physicians could 1. Gold MR, Hurley R, Lake T, Ensor T, Berenson R. A national survey of the arrangements managed-care plans make with physicians. N EngI J Med 1995; accept capitation for individuals with CAD with an 333:1678-1683. upper limit threshold of $10,000. The physicians 2. Robinson JC, Casalino LP. The growth of medical groups paid through capitation in California. N Engl J Med 1995;333:1684-1687. would be financially responsible for the entire cost of 3. Simon CJ, Emmons DW. Physician earnings at risk: an examination of the 78% of treatment (from Figure 2) which cost capitated contracts. Health Ajf2air.s 1997;16:120-126. TS, Grumbach K. Capitation or decapitation: Keeping your head <$lO,OOOas well as the first $10,000 of the remaining 4.in Bodenheimer changing times. JAMA 1996;276: 1025-1031. 22% of costs. The reinsurer would be financially re- 5. Newhouse JP. Patients at risk: health reform and risk adjustment. Health sponsible for the costs >$lO,OOOfor these 22% higher Aflairs 1994;13’132-.146. 6. Newhouse JP. Reimbursing health plans and health providers: efficiency in costs. Given the very high cost of these expenses the production versus selection. J Econ Lit 1996;34:1236-1263. 7. Neff JM, Anderson. Protecting children with chronic illness in a competitive cost of reinsurance would be quite high. marketplace. JAMA 1995;274:1866-1869. limitation of Medicare data: Using Medicare claims 8. Andrew IS, Anderson GF, Han C, Neff JM. Pediatric carve outs. The use of conditions as risk adjusters in capitated payment systems. Arch data to examine expenditures by diagnosis has several disease-specific Pediatr Ad&x Med 1997;15 1:236-242. limitations. Medicare data were not designed for clin- 9. Physician Payment Review Commission. Annual Report to Congress 1994. DC, 1994:135-158. ical analysis; thus, much of the data are not as com- Washington, 10. Weiner JP, Dobson A, Maxwell SL, Coleman K, Starfield BH, Anderson GF. plete as in a clinical database.19For instance, only Risk-adjusted Medicare capitation rates using ambulatory and inpatient diagcrude data on severity of diagnosis is directly in- noses. Health Care Finan Rev 1996;17:77-99. 11. Ellis RP, Pope GC, Iezzoni LI, et al. Diagnosis-based risk adjustment for eluded. Second, miscoding has been well document- Medicare capitation payments. Health Care Finan Rev 1996;17:101-128. ed.14Jo,21Third, cross-sectional data, although inter- 12. Kmnick R, Dreyfus T, Lee L, Zhou Z. Diagnostic risk adjustment for Medicaid: the disability payment system. Health Care Finan Rev 1996;17:7-33. esting, only provide an initial picture. Expenditures in 13. Iversen LH, Oberg CN, Polich CL. The availability of long-term care services 1 year may not accurately reflect costs in subsequent for Medicare beneficiaries in health maintenance organizations. Med Care 1988; years.22 Fourth, unlike age and gender, parameters 26:918-925. 14. Van Walraven C, Wang B, Ugnat A, Naylor CD. False-positive coding for based on reported diagnostic codes could be manipu- acute myocardial infarction on hospital discharge records: chart audit results from centre. Can J Cardiol 1990;6:383-386. lated by some providers. 23 Fifth, Medicare claims do a15.tertiary Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of not provide data for nonreimbursed items such as classifying prognostic comorbidity in longitudinal studies: development and J Chron Dis 1993;40:373-383, pharmaceutical costs, eyeglass and/or dental care, cer- validation. 16. Deyo RA, Cherkin DC, Ciol MA. Adapting a clinical comorbidity index for tain elective procedures, and long-term care that man- use with ICD-9.CM administrative databases. J Clin Epidemiol 1992;45:613aged care organization may cover in their benefit plan. 619. 17. Epstein AM, Cumella EJ. Using predictors of medical utilization to adjust Finally, analysis of data on the elderly may not be rates. Health Care Financ Rev 1988;10:51-69. 18. Anderson G, Steinberg EP, Whittle J, Powe NR, Antebi S, Herbert R. applicable to other age groups. Administrative data- Development of clinical and economic prognosis from Medicare claims data. bases with better clinical measures are being devel- JAMA 1990;263:967-972. oped,19but they are unlikely to contain the large and 19. Jollis JG, Ancukiewicz M, DeLong ER, Pryor DB, Muhlbaier LH, Mark DB. Discordance of databases designed for claims payment versus clinical informabroadly representative number of people with compre- tion systems: implications for outcomes research. Ann Intern Med 1993;119:84& hensive financial data that Medicare has in its data- 850. 20. Whittle J, Steinberg EP, Anderson GF, Herbert R. Accuracy of Medicare base. claims data for estimation of cancer incidence and resection rates among elderly Thus, managed care organizations may set rates Americans. Med Care 1991;29:1226-1236. lezzoni LI, Bumside S, Sickles L, Moskowitz MA, Sawitz E, Levine PA. based on these large administrative databases,such as 21. Coding of acute myocardial infarction: clinical and policy implications. Ann Medicare files, becausecurrently no other comprehen- Intern Med 1988;109:745-751. Welch WP. Regression toward the mean in medical care costs. Implications sive data exist. Unfortunately, these types of data are 22. for biased selection in health maintenance organizations. Med Care 1985;23: usually not available to physicians when they sign 1234-1241. 23. Assaf AR, LaPane KL, McKenney JL, Carleton RA. Possible influence of the contracts. Understanding the implications of capita- prospective payment system on the assignment of discharge diagnoses for coretion is critical for cardiologists. nary heart disease. N Engl J Med 1993;329:931-935.
1182
THE AMERICAN JOURNAL OF CARDIOLOGY@
VOL. 82
NOVEMBER
15, 1998
annual expenditures for the general Medicare pop& tion ($6,944 vs $3,247). Among individuals with CAD, mean ex nditures of subgroups defined by both age and gen cf= er ranged from $6,205 to $7,724. In comparison, stratifying by measures of severity and comorbidity identified subgroups with lower and higher mean ex nditures, producing a range of $1,702 to $19,959. Sur stantial variation of expenditures for individuals within subgroups defined by severity and comorbidity remained, with few patients having substantially higher expenditures than the rest. When capitation rates are set with the use of demographic factors alone, patients may be subiected to risk selection and physicians to financial loss. Using clinical measures may decrease the incentive for patient risk selection, but substantial financial risk to physicians would remain, because of a relatively few patients with high expenditures (or costs). 01998 by Excerpta Medica, Inc. (Am J Cardiol 1998;82:1178-1182)
apitation is becoming a more common payment method not only to managed care organizations C but also to physician groups.lJ One-third of United
payments, more comprehensive risk adjustment systems classify individuals into an expected cost category based on demographic and clinical factors.lO-l* We examined 1992 Medicare expenditures for individuals with coronary artery disease (CAD) to compare mean expenditures and variation in expenditures of subgroups defined by demographic factors with those defined by severity and comorbidity, which are particularly relevant to small physician groups who have limited ability to spread financial risk.
States (US) physicians had at least 1 capitated contract in 1995.3 Although capitation at the physician level has historically been reserved for primary care physicians, certain specialists, including cardiologists, are now entering into capitated arrangements as well.’ In theory, capitation provides a financial incentive for physicians to deliver care more efficientlye Unfortunately, capitation provides a disincentive to select individuals with chronic illnesses who have high expected costs.536Adjusting capitated rates to more accurately reflect the expected cost of care has been proposed to protect sicker patients from risk selection, as well as to protect physicians with capitation contracts who care for them from excessive financial 10ss.~,~,s In 1 method of risk adjustment, physicians receive a separate, usually higher, capitation payment for each individual with an identified chronic illness that is expensive to treat.” To further adjust capitation From the Departmen% of Epidemiology, Departmen. of Medicine, and Hea,th PO icy and Mona ement, Tie Johns Hopkins Medical Institutions, Bal-imore, Marylan 3 This study was supported by the Commonwealth Fund Grant #96169, New York, ‘\lew YOK. Manuscript received February 9, 1998; revised manuscripr rece’ved and accepted JLne 19, 1998. Address for rep-ints: Robert 1. McNamara, MD, MHS, School of Hygiene and Public Health, Johns Yopkins Jn’vexity, 65 %orth Wolfe Sheet, Room 6009, Baltimore, Mayland 21205. E-mail: [email protected].
1178
01998 by Excerpta Medica, All rights reserved.
Inc.
METHODS Patient selection: Data were obtained from standard research files from the Health Care Financing Administration, consisting of a 5% nationally random sample of Medicare beneficiaries. The enrollment file contains demographic information on each Medicare beneficiary, and the claims file contains expenditures and utilization at the individual beneficiary level for all Medicare covered services. Medicare Part A claims included hospital inpatient and hospital outpatient claims. Medicare Part B included physician inpatient and outpatient services as well as supplier services. International Classification of Disease (ICD-9) codes for services rendered were included with the expenditure data. We combined the enrollment and claims files to develop individual records for each Medicare beneficiary in our sample. We excluded Medicare beneficiaries <65 years of age, who typically have permanent disability or endstage renal disease and may confound a resource anal0002.9149/98/S 19.00 PII SOOO2-9149(98)00602-X
CAD (n = 56,820), without any Medicare Part B claims involving CAD (n = 4,278), and with claims of zero dollars (n = 3,042) were excluded. Thus, the General Medicare Coronary Artery Disease final CAD- data set included claims from 250,514 % % beneficiaries. Age b) Persons with CAD were classified further into 1 of 65-69 28.0 21.0 5 specific diagnoses based upon the 5 3-digit codes 25.2 23.7 70-74 (410 to 414) that were used because the fourth and 75-79 20.5 22.5 fifth digits are often missing. If a person had two 410 14.1 17.0 80-84 12.3 15.8 85+ codes, regardless of other CAD codes, they were Gender classified into the 410 subgroup. If a person did not 61.8 54.5 Women have two 410 codes but did have two 411 codes, they 32.2 45.5 Men were classified into the 411 subgroup. The other 3 Charlson Index subgroups were similarly defined in this hierarchical 44.4 24.7 0 1 30.0 28.8 fashion. 2 14.7 21.7 Comorbidity definition: The Charlson index is a 3+ 10.9 24.8 method of classifying individuals by comorbid conditions that alter the risk of mortality.rs Each of 17 comorbid conditions is assigned a value from 1 to 6. For $6,W instance, chronic obstructive pulmo$7,000 nary disease is assigned a weight of 1, and metastatic solid tumor is as$6,000 signed a weight of 6. The Charlson index for an individual is the sum of Mean $5,000 these values for the comorbidities of Expenditures $4,000 that individual. Deyo and col1992 leaguesI6 adapted this index, which $3,000 i was based on review of medical $2,000 records, for use with administrative databases.We applied this method to $1,000 the Medicare files to provide a Charlson index score for each beneficiary. $0 : For simplicity, and to achieve comGeneral Medicare Individuals with parable numbers in each group, we Population Coronary Artery Disease chose to combine all the Charlson indexes of 23 into 1 group. FIGURE 1. Mean expenditures for he general Medicare population (n = 1,221,615) Statistical analysis: The basic anaversus individuals with CAD (n = 250,514). lytic strategy was to compare mean expenditures and variation in expenysis. Beneficiaries living outside the US, those who ditures between different subgroups of Medicare bendid not have Part A and Part B coverage, and those eficiaries. Because statistically significant differences who were enrolled in a managed care plan were not may not be clinically significant when numbers are included due to incomplete utilization information. large, the comparisons were illustrated graphically. Thus, our final data set included 1,221,615 Medicare beneficiaries >65 years old living in the US with RESULTS Individuals with CAD who met our criteria (n = fee-for-service coverage in 1992. The claims files were further refined by excluding 250,514) were older, more likely men, and had more expenditure data from skilled nursing homes, hospice comorbidity than the general Medicare population care, and home health care. Although some managed (n = 1,221,615) (Table I). Per capita annual expendicare organizations cover these services, they are often tures by the Medicare program for individuals with paid by Medicare on a fee-for-service basis. In addi- CAD were more than double those of the general tion, how plans define and structure these benefits Medicare population ($6,944 vs $3,247) (Figure 1). varies across plans. ‘3 Diagnostic and procedural codes Such a difference may provide a financial disincentive from all other covered services were reviewed. to enroll individuals with CAD unless payment rates Coronary artery disease diagnosis definition: This reflect the higher expected cost. Therefore, establishment of a separatecapitation rate for individuals with was identified by the ICD-9 Clinical Modification (ICD-9-CM) codes 410 to 414. Because CAD fre- CAD is an attractive proposal to minimize this disinquently is misclassified in claims databasesi benefi- centive. The shape of the curve illustrating the distribution ciaries with CAD were defined as having a CAD code at least twice, with 2 1 claim from Medicare Part B. of expenditures of individuals with CAD was similar Beneficiaries without 2 separate claims involving to that of the general Medicare population (Figure 2). TABLE I Comparison Population
Between the General and Individuals With Coronary
Medicare Artery Disease
CORONARY ARTERY DISEASE/CAPITATION
FOR CORONARY ARTERY DISEASE
1179
factors that predict costs better, the incentive to select individuals with low expected costs and to avoid indi$50,000 -viduals with high expected costs would remain. $40,000 -In addition to demographic inforExpenditures mation, some clinical information is $30,000 -1992 available in Medicare claims files. Clinical information from the previ$20,000 -ous year or years could be used prospectively to set the subsequent $10,000 -year’s capitation rates. We used diagnostic codes available on Medicare $0 claims files to stratify individuals 100% 80% 40% 80% 0% 20% with CAD. As a measure of severity, Percentile of Population persons were stratified by 3-digit ICD-9 diagnostic codes for CAD. For FIGURE 2. Distribution of expenditures by percentile for the general Medicare popuinstance, acute or recent myocardial lation (n = 1,221,615) and for individuals with CAD (n = 250,514). infarction is ICD-9 code 410. In addition, we stratified the beneficiaries based on the Charlson index, a wellvalidated indicator of comorbid$8,000 ity.15J6 Mean expenditures of persons $7,000 differed substantially when stratified by these measurements of severity $6,000 and comorbidity (Figure 4). Persons with acute myocardial infarction and a Exp~~~rures $5’ooo Charlson index of ~3 (high comor1992 $4,000 bidity) had the highest mean expenditures ($19,959) and those with un$3,000 specified atherosclerosis and no co$2,000 morbidities had the lowest ($1,702). Variation in expenditures: Adjust$1,000 ing the capitation rate only to reflect $0 mean expenditures within subgroups, however, may not protect the provider 75-79 60-64 as+ 65-69 70-74 from variation in cost. We examined how the measurementsof severity and Age comorbidity affected variation in exFIGURE 3. Mean expenditures of subgroups of individuals with CAD stratified by age penditures by illustrating mean expenand gender. For women (white bars): 65 to 69 years old, n = 23,473; 70 to 74 ditures by percentile for each meayears old, n = 28,874; 75 to 79 years old, n = 30,314, 80 to 84 years old, n = sure. Stratifying these measurements 25,860; and 285 years old, n = 28,117. For men /black bars): 65 to 69 years old, did little to change the shape of the n = 29,036; 70 to 74 years old, n = 30,552; 75 to 79 years old, n = 26,026; 80 to 84 years old, n = 16,707; and ~85 years old, n = 11,555. distribution of expenditures (Figure 5 compared with Figure 2). A few individuals continued to have substanAlthough the expenditures of most people were below tially higher expenditures in each subgroup. Thus, the mean for each group, the expenditures for a few rates that account for severity and comorbidity in this were considerably higher. Thus, establishing a sepa- way would not protect the physician who accepts rate capitation rate for individuals with CAD would capitation from substantial financial loss due to a few not eliminate the financial risk to providers because of patients with high cost expenses. a relatively few patient with high cost expenses. Factors affecting cost: Because age and gender are DISCUSSION Managed care organizations increasingly are using currently used by Medicare to set rates for beneficiaries enrolling in managed care organizations and by capitation as a means to compensatephysicians. Large some managed care organizations to set capitation administrative databases, such as Medicare expendirates to physicians, we examined mean expenditures ture files, are likely to be a source of data in setting of individuals with CAD stratified by these factors. payment rates. Currently, Medicare pays managed However, this method did not substantially differen- care organizations a per capita rate of 95% of the tiate mean expenditures among groups. Mean expen- expected costs for the average fee-for-service benefiditures ranged from $6,205 to only $7,724 (Figure 3). ciaryr” with adjustments for age, gender, welfare staIf payers or capitated providers know of additional tus, institutional status, and eligibility status of bene$80,000 -
ii80
THE AMERICAN JOURNAL OF CARDIOLOGY@
VOL. 82
NOVEMBER
15, 1998
FIGURE4. Mean expenditures of subgroups of individuals with CAD stratified by measures of severity and comorbidity. 4 10 = acute or recent myocardial infarction (n = 30,461); 411 = unstable angina (n = 26,587); 412 = prior myocardiaf infarction (n = 8,004); 413 = stable angina (n = 46,289); and 414 = unspecified atherosclerosis (n = 139,173). Charlson index of 0 (n = 71,748), Charlson index of 1 (n = 74,797), Charlson index of 2 (n = 52,31 l), and Char&on index ~3 (n = 50,658).
$80,000
T 410
$70,000
411
$60,000 Expenditures 1992
J-Digit ICD-9 412 Code 413 414
$50,000 $40,000 $30,000 $20,000 $10,000 $0
0%
20%
40%
60%
80%
100%
Percentile of Population A $100,000
T
I 3+
$80,000 ! II
0%
20%
40%
60%
80%
2 Charlson Index
100%
Percentile of Population B FIGURE 5. Distribution of expenditures by percentile for individuals with CAD stratified by /A) severity and (Bj comorbidity. See Figure 4 for numbers in each group.
ficiaries.r7 Capitation rates based on expected costs of the average Medicare beneficiary discourages managed care organizations and physicians from caring for people with chronic illnesses such as CAD due to the higher expected costs of this population. Setting different capitation rates based upon an individual’s principal diagnoses is 1 type of risk-adjustment method. Age and gender, 2 of the factors currently used to set capitation rates for Medicare beneficiaries enrolling in managed care organizations, did not identify subgroups of individuals with CAD with very high or very low mean expenditures. However, the other factors available in large administrative databases that we examined-measurements of severity and comorbidity-did identify subgroups with high and low mean expenditures. Thus, the use of demographic factors would do little to decreasethe financial incentive to avoid individuals with high expected cost, but the use of measurements of severity and comorbidity likely would decreasethis incentive. Variation in cost is particularly relevant to small cardiology groups with limited ability to spread financial risk. With 10% of the general Medicare population accounting for 70% of the total expendituresI a few patients with high costs could lead to financial loss for providers accepting capitation. Unfortunately, our results indicate that disregarding CAD does not limit variation in expenditures either as a whole group or stratified by measurements of severity or comorbidity (Figures 2 and 5). Thus, the use of these measurements of severity and comorbidity would not protect physicians who accept capitation from substantial financial loss due to a few patients with high cost expenses. Reinsurance: Alternative risk-adjustment methods other than diagnostic carveouts are available. Reinsurance, or stop-loss insurance, is one of the most common and covers unexpected losses from extremely high cost expenses.Basically, the individual or group accepting a capitated payment is responsible for annual costs only up to a predefined level. Above that level, the reinsurer assumestotal or partial financial re-
CORONARY ARTERY DISEASE/CAPITATION
FOR CORONARY ARTERY DISEASE
118 1
sponsibility. For instance, a group of physicians could 1. Gold MR, Hurley R, Lake T, Ensor T, Berenson R. A national survey of the arrangements managed-care plans make with physicians. N EngI J Med 1995; accept capitation for individuals with CAD with an 333:1678-1683. upper limit threshold of $10,000. The physicians 2. Robinson JC, Casalino LP. The growth of medical groups paid through capitation in California. N Engl J Med 1995;333:1684-1687. would be financially responsible for the entire cost of 3. Simon CJ, Emmons DW. Physician earnings at risk: an examination of the 78% of treatment (from Figure 2) which cost capitated contracts. Health Ajf2air.s 1997;16:120-126. TS, Grumbach K. Capitation or decapitation: Keeping your head <$lO,OOOas well as the first $10,000 of the remaining 4.in Bodenheimer changing times. JAMA 1996;276: 1025-1031. 22% of costs. The reinsurer would be financially re- 5. Newhouse JP. Patients at risk: health reform and risk adjustment. Health sponsible for the costs >$lO,OOOfor these 22% higher Aflairs 1994;13’132-.146. 6. Newhouse JP. Reimbursing health plans and health providers: efficiency in costs. Given the very high cost of these expenses the production versus selection. J Econ Lit 1996;34:1236-1263. 7. Neff JM, Anderson. Protecting children with chronic illness in a competitive cost of reinsurance would be quite high. marketplace. JAMA 1995;274:1866-1869. limitation of Medicare data: Using Medicare claims 8. Andrew IS, Anderson GF, Han C, Neff JM. Pediatric carve outs. The use of conditions as risk adjusters in capitated payment systems. Arch data to examine expenditures by diagnosis has several disease-specific Pediatr Ad&x Med 1997;15 1:236-242. limitations. Medicare data were not designed for clin- 9. Physician Payment Review Commission. Annual Report to Congress 1994. DC, 1994:135-158. ical analysis; thus, much of the data are not as com- Washington, 10. Weiner JP, Dobson A, Maxwell SL, Coleman K, Starfield BH, Anderson GF. plete as in a clinical database.19For instance, only Risk-adjusted Medicare capitation rates using ambulatory and inpatient diagcrude data on severity of diagnosis is directly in- noses. Health Care Finan Rev 1996;17:77-99. 11. Ellis RP, Pope GC, Iezzoni LI, et al. Diagnosis-based risk adjustment for eluded. Second, miscoding has been well document- Medicare capitation payments. Health Care Finan Rev 1996;17:101-128. ed.14Jo,21Third, cross-sectional data, although inter- 12. Kmnick R, Dreyfus T, Lee L, Zhou Z. Diagnostic risk adjustment for Medicaid: the disability payment system. Health Care Finan Rev 1996;17:7-33. esting, only provide an initial picture. Expenditures in 13. Iversen LH, Oberg CN, Polich CL. The availability of long-term care services 1 year may not accurately reflect costs in subsequent for Medicare beneficiaries in health maintenance organizations. Med Care 1988; years.22 Fourth, unlike age and gender, parameters 26:918-925. 14. Van Walraven C, Wang B, Ugnat A, Naylor CD. False-positive coding for based on reported diagnostic codes could be manipu- acute myocardial infarction on hospital discharge records: chart audit results from centre. Can J Cardiol 1990;6:383-386. lated by some providers. 23 Fifth, Medicare claims do a15.tertiary Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of not provide data for nonreimbursed items such as classifying prognostic comorbidity in longitudinal studies: development and J Chron Dis 1993;40:373-383, pharmaceutical costs, eyeglass and/or dental care, cer- validation. 16. Deyo RA, Cherkin DC, Ciol MA. Adapting a clinical comorbidity index for tain elective procedures, and long-term care that man- use with ICD-9.CM administrative databases. J Clin Epidemiol 1992;45:613aged care organization may cover in their benefit plan. 619. 17. Epstein AM, Cumella EJ. Using predictors of medical utilization to adjust Finally, analysis of data on the elderly may not be rates. Health Care Financ Rev 1988;10:51-69. 18. Anderson G, Steinberg EP, Whittle J, Powe NR, Antebi S, Herbert R. applicable to other age groups. Administrative data- Development of clinical and economic prognosis from Medicare claims data. bases with better clinical measures are being devel- JAMA 1990;263:967-972. oped,19but they are unlikely to contain the large and 19. Jollis JG, Ancukiewicz M, DeLong ER, Pryor DB, Muhlbaier LH, Mark DB. Discordance of databases designed for claims payment versus clinical informabroadly representative number of people with compre- tion systems: implications for outcomes research. Ann Intern Med 1993;119:84& hensive financial data that Medicare has in its data- 850. 20. Whittle J, Steinberg EP, Anderson GF, Herbert R. Accuracy of Medicare base. claims data for estimation of cancer incidence and resection rates among elderly Thus, managed care organizations may set rates Americans. Med Care 1991;29:1226-1236. lezzoni LI, Bumside S, Sickles L, Moskowitz MA, Sawitz E, Levine PA. based on these large administrative databases,such as 21. Coding of acute myocardial infarction: clinical and policy implications. Ann Medicare files, becausecurrently no other comprehen- Intern Med 1988;109:745-751. Welch WP. Regression toward the mean in medical care costs. Implications sive data exist. Unfortunately, these types of data are 22. for biased selection in health maintenance organizations. Med Care 1985;23: usually not available to physicians when they sign 1234-1241. 23. Assaf AR, LaPane KL, McKenney JL, Carleton RA. Possible influence of the contracts. Understanding the implications of capita- prospective payment system on the assignment of discharge diagnoses for coretion is critical for cardiologists. nary heart disease. N Engl J Med 1993;329:931-935.
1182
THE AMERICAN JOURNAL OF CARDIOLOGY@
VOL. 82
NOVEMBER
15, 1998