This Article Abstract Submit a response to this article Services E-mail this article to a friend Alert me to new issues of the journal

V A R I A T I O N S :
C A L I F O R N I A
 
16 November 2005  
  

Evaluating The Efficiency Of 
California Providers In Caring 
For Patients With Chronic Illnesses

Important opportunities exist for hospitals, in California
and elsewhere, to improve the efficiency of their
chronic illness care.


by John E. Wennberg, Elliott S. Fisher, Laurence Baker,
Sandra M. Sharp, and Kristen K. Bronner


ABSTRACT:

In this paper we compare the relative efficiency of health care providers in managing patients with severe chronic illnesses over fixed periods of time. To minimize the contribution of differences in severity of illness to differences in care management, we evaluate performance over fixed intervals prior to death for patients who died during a five-year period, 1999–2003. Medicare spending, hospital bed and full-time equivalent (FTE) physician inputs, and utilization varied extensively between regions, among hospitals located within a given region, and among hospitals belonging to a given hospital system. The data point to important opportunities to improve efficiency.

Widespread concern about the quality and cost of medical care has spurred public- and private-sector efforts to measure the performance of individual physicians, medical groups, hospitals, nursing homes, and health plans. Currently available performance measures, however, are limited in their scope. Most focus on the technical processes of care—in particular, the underuse of effective care, such as indicated screening tests in diabetics. Others include reports on patients’ experiences of care, structural indicators, and outcome measures such as risk-adjusted mortality following acute myocardial infarction (AMI). Virtually none addresses the amount of money spent, resources used, and services provided per person over fixed intervals of time for similarly ill patients. In the absence of such population-based measures, payers’ efforts to improve efficiency have been based on the severity-adjusted unit price of services—for example, Medicare’s diagnosis-related group (DRG)–based payable inpatient stay charges. This is unfortunate, because most of the more than twofold difference across U.S. regions in Medicare per capita spending is due to differences in resource inputs (physician and hospital resources per capita) and the associated use of services—in particular, the per person frequency of physician visits and use of the hospital as a site of care for patients with chronic illnesses.¹

Evidence at the regional level suggests that greater spending, more resource inputs, and more frequent use of hospitals and physician services are not associated with better performance on technical measures of the processes of care.² Neither are they associated with improvements in survival, functional status, or satisfaction with care.³ A related study found no evidence that more care results in better quality or survival among patients cared for by academic medical centers.⁴ This research suggests that regions or hospitals with low spending, resource inputs, and utilization are more efficient in managing chronic illnesses than those providing more care, and their longitudinal performance measures could serve as benchmarks for evaluating relative efficiency, provided that spending, resource input, and utilization measures are adequately adjusted for case-mix and disease severity and that available measures of quality are satisfactory.

We recently described a method of using Medicare claims data to create longitudinal performance measures for cohorts of patients with severe chronic illnesses, according to the hospital (and associated physicians) from which they receive most of their care.⁵ The measures are population-based and include Medicare spending, resource inputs (hospital and intensive care unit, or ICU, beds and full-time equivalent, or FTE, physician labor) and frequency of use of hospitals, ICUs, and physician services. To minimize the possibility that variations in performance could be explained by differences among patients, we adjusted the measures for age, sex, race, and relative frequency of specific chronic illnesses. Moreover, the measures were restricted to six-month intervals during the two-year period prior to death, which ensured that the populations assigned to hospitals had identical prognoses.

Publication of this paper coincides with the public release of these measures on the Dartmouth Atlas of Health Care Web site, www.dartmouthatlas.org, for California hospitals during the five-year period 1999–2003. In this paper we provide an overview of our methods; illustrate why longitudinal population-based, hospital-specific measures are important; and use our data to evaluate the relative efficiency of selected California regions, hospitals, and hospital systems.

Study Populations And Methods

The database. The primary database is four research files from the Centers for Medicare and Medicaid Services (CMS) for traditional (fee-for-service) Medicare: the Medicare Provider Analysis and Review (MEDPAR) file; the Medicare Part B file (a 20 percent sample); the Outpatient file; and the Denominator file that contains demographic data and date of death.

Study populations. We used claims data for Medicare beneficiaries who died during 1999–2003 and who were hospitalized at least once during the last two years of life. We further restricted the analysis to patients who had one or more of twelve chronic illnesses associated with a high probability of death.⁶ Claims data were used to assign each patient to the hospital used most often during the last two years of life. In the case of a tie, patients were assigned to the hospital associated with the discharge closest to the date of death. The assignment of patients to hospitals is robust because most patients with serious chronic illnesses use the same hospital for most of their care, as shown elsewhere.⁷ Population-based rates, however, were calculated based on the total experience of the population over the given period of time, not only from the care received at the assigned hospital or physicians associated with that hospital. The regional analyses include patients who were residents of a given region at the time of death; the hospital-specific analyses are restricted to hospitals with 400 or more deaths during the five-year study period. Data for all hospitals with eighty or more deaths during the study period are available on the Dartmouth Atlas Web site.

Measures of spending. Inpatient reimbursements were calculated by summing Medicare reimbursements from the MEDPAR record; they reflect total reimbursements, including indirect costs for medical education, disproportionate-share hospital (DSH) payments, and outlier payments. Part B payments were for all services included in the Part B Physician Supplier file. Inpatient reimbursements and Part B payments were measured as spending per decedent.

Measures of resource inputs. Measures of resource inputs include physician labor, hospital beds, ICU beds, and Medicare spending (reimbursements), presented here as summary measures over the last two years of life. Bed input rates were calculated by summing patient days and dividing by 365. Physician labor inputs were measured by summing the specialty-specific work relative value units (RVUs) and dividing by the average annual number of work RVUs produced by that specialty. The measure is used to estimate the standardized FTE physician labor input. Both bed and FTE physician resources are expressed as inputs per 1,000 decedents.

Measures of utilization. We calculated hospital days, ICU days, and physician visits (overall and separately for primary care physicians and medical specialists) for each patient during the last six months of life. These measures of utilization are traditional epidemiological, population-based rates of events occurring over a designated period of time. Although utilization rates were calculated on the total experience of the cohort, the proportion of total care provided by the assigned hospital is high, so the variations in utilization among hospital cohorts primarily reflect the associated physicians’ clinical choices.

Quality indicators. Two claims-based quality-of-care measures were used. The percentage of patients seeing ten or more physicians is a measure of the propensity to refer patients. The inverse correlation between this measure and the use of effective care (such as beta-blockers after heart attacks) suggests that when many physicians become involved, care management becomes less effective. The percentage of deaths occurring during a hospitalization that involved one or more stays in an ICU is an indicator of the aggressiveness with which terminal patients were treated. In light of the evidence that more-aggressive care in managing patient populations with chronic illness does not lead to longer length or improved quality of life, higher scores on this measure can be viewed as an indicator of unnecessarily intensive end-of-life care.

We also report quality measures regarding the processes of care—specifically, the underuse of effective care, derived from the consensus measure set of the Hospital Quality Alliance (HQA), which is the first initiative to routinely report data on U.S. hospitals nationally. Data are posted on the CMS Web site, www.cms.hhs.gov. We provide summary scores on five measures for managing acute myocardial infarction (AMI); two for congestive heart failure (CHF); and three for pneumonia, for all reporting hospitals located within each Hospital Referral Region (HRR). For individual hospitals, summary scores are based on measures for which there are twenty-five or more eligible patients.⁸

To examine patients’ assessment of the quality of hospital care, we used results from the 2004 California Hospital Experience Survey. The survey asked 36,000 patients treated at 200 California hospitals to rate their hospital experience along several dimensions, including respect for patients’ preferences, coordination of care, and the involvement of family and friends. Scores on each dimension were then combined to develop an overall patient experience score, and hospitals were rated as below average, average, or above average, depending on how their scores compared with the California statewide average. For individual hospitals, we report this rating; for regions, we report the percentage of hospitals in the region that patients rated above and below average.⁹

Statistical methods. We compared measures of Medicare spending, resource inputs, utilization, and quality at fixed intervals prior to death among geographic regions and among hospitals in California. All spending, resource input, and utilization measures were further adjusted for differences in age, sex, race, and the relative predominance of the twelve chronic conditions, using ordinary least squares regression for Medicare spending variables and overdispersed Poisson regression models for all other variables; 95 percent confidence limits were calculated for all variables. The HQA technical process-quality-of-care measures were not adjusted for differences in case-mix among hospitals, because they are specifically restricted to patients who are eligible for the specific treatment and do not, therefore, need adjustment. The patient-experience measures were adjusted for case-mix differences, as described elsewhere.¹⁰

Study Results

Importance of longitudinal population-based, hospital-specific performance measures. Per capita Medicare spending. The per capita cost of medical care over time is a function of the per capita volume and the per unit price of care: cost per capita = units of care per capita x cost per unit of care. In the absence of population-based information on volume, the evaluation of efficiency has had to rely on cost per unit of care. The importance of taking both per capita volume and the unit price of care into account is underscored in Exhibit 1. It examines the association between reimbursements per person over time for inpatient care and the volume of hospital care provided during the same time period, measured by hospital days per person. Among the 226 California hospitals with more than 400 deaths, per person spending in the last two years of life ranged from less than $20,000 to almost $90,000. Two-thirds of the variation in per person spending, measured by the R² statistic, was associated with variation in hospital days per person. Only 39 percent of the variation was associated with price per day (data not shown).¹¹

Hospital level of population aggregation. Exhibit 2 illustrates the importance of aggregating chronically ill patient populations according to the hospital they most often use. Age, race, sex, socioeconomic status (SES), and medical condition were associated with hospitalization rates to varying degrees; on average, during the last six months of life, younger Medicare patients spent 1.49 times more days in the hospital than older patients; blacks, 1.20 times more days than nonblacks; males, 1.07 times more days than females; patients with lower SES (indicated by Medicaid buy-in), 1.06 times more days than all others; and CHF patients, slightly more days (1.02 times) than cancer patients. However, the range of variation according to the hospital most often used was much greater for each subgroup. For example, although younger patients were hospitalized 1.49 times more often than older patients, patient day rates for both groups varied more than threefold among the hospitals; moreover, the rates for older and younger patients were highly correlated (R² = 0.78), which indicates that the relative variation associated with hospital was consistent across age categories. Although, on average, blacks spent 1.20 times more days in the hospital, patient day rates for both black and nonblack patients varied greatly across hospitals. Blacks using the hospital with the highest usage of days spent 2.66 times more days as inpatients than did those in the hospital with the lowest patient day rate; for nonblacks, the ratio was 2.40. Again, hospitals with high use rates for blacks had high use rates for nonblacks (and vice versa), as evidenced by the high R² (0.79). Similar patterns are evident for subgroups defined by sex, SES, and condition.

Exhibit 2 also provides evidence that the hospital effect is independent of severity of illness. During the last six months of life (when acuity of illness for the cohort was extreme), patients spent on average 5.32 times as many days in the hospital as they did during the nineteen to twenty-four months prior to death (when acuity was less). However, during each interval of time prior to death, days in the hospital varied almost fourfold, and the rates were highly correlated (R² = 0.74). Similar effects were seen for physician visits (data not shown).

Benchmarking relative efficiency in managing chronic illness: California regions. Here we compare Medicare spending, resource inputs, and quality measures for chronically ill Medicare decedents in four California regions: Los Angeles, San Francisco, San Diego, and Sacramento (Exhibit 3). On the basis of its lower spending, lower resource inputs, and utilization rates and its relatively satisfactory quality measures, we selected the Sacramento region as the regional benchmark for relative efficiency.

Medicare spending. Spending for physician services (Part B payments) during the last two years of life was 1.74 times greater in Los Angeles; 1.09 times greater in San Francisco; and 1.29 times greater in San Diego, compared with Sacramento. Inpatient reimbursements were 1.67, 1.39, and 1.16 times greater than the Sacramento benchmark, respectively.¹² By contrast, inpatient reimbursements per day in the hospital were about the same in Los Angeles, San Diego, and Sacramento, while in San Francisco they exceeded those of Sacramento by a factor of 1.16.

Resource inputs. Providers serving Sacramento used consistently fewer resource inputs for hospital and ICU beds and FTE physicians per 1,000 than the other three regions. The greatest contrast was with Los Angeles, where providers used 1.61 times more hospital beds, 2.28 times more ICU beds, and 1.89 times more physicians in caring for chronically ill patients during the last two years of life.

Utilization. During the last six months of life, patients living in Los Angeles spent 1.62 times more days in the hospital and 2.31 times more days in ICUs, and they visited their physicians 2.34 times more often than their counterparts in Sacramento. Rates for these services were also higher in San Francisco and San Diego than in Sacramento. Residents of Los Angeles were much more likely to experience a “high-tech” death: 33 percent of decedents died during a hospitalization that included a stay in the ICU, compared with 19 percent of decedents in Sacramento. During the last six months of life, those in Los Angeles were much more likely to be referred to other physicians (Exhibit 3).

Quality. The CMS quality measures indicated relatively lower scores for hospitals serving Los Angeles and San Diego than for Sacramento and San Francisco; 57 percent of surveyed hospitals in Los Angeles were rated as below average by the patients using them, compared with 13 percent and 9 percent of hospitals in Sacramento and San Francisco, respectively.

Benchmarking relative efficiency in managing chronic illness: Los Angeles hospitals. Although the Los Angeles region is noted for high-intensity care, there was considerable variation among its hospitals in their management of patients with chronic illnesses and in quality of care (Exhibits 4, 5, and 6). However, per person Medicare spending, resource inputs, and utilization in every Los Angeles hospital listed in the exhibits exceeded the Sacramento regional benchmark, with the exception of the inpatient unit price variable (average reimbursements per day in the hospital).

Medicare spending. Among the twenty-eight hospitals listed in Exhibit 4, Medicare spending during the last two years of life varied by a factor of 2.76, from $38,567 at Foothill Presbyterian Hospital to $106,254 at Garfield Medical Center. Although every Los Angeles hospital exceeded the Sacramento benchmark for inpatient reimbursements per decedent, inpatient unit price exceeded the benchmark in only nine of the twenty-eight hospitals. The spending profiles of these hospitals further illustrate the importance of taking both volume and price into account. For example, inpatient spending rates at the most costly hospital, Garfield Medical Center, were $88,601, which exceeded the Sacramento benchmark ($26,048) by a factor of 3.40, achieved because reimbursements per day ($2,306) were 1.68 times higher than the Sacramento benchmark ($1,370) and patient day rates at Garfield Medical Center (23.4, from Exhibit 5) were 2.02 times higher than the Sacramento benchmark (11.1, Exhibit 5). By contrast, price per day at St. Mary Medical Center was lower than the benchmark (0.91), even though per capita spending was 1.78 times higher because of high volume; during the last two years of life, patient days per decedent were 1.96 times those of Sacramento.

Resource inputs. Although Los Angeles hospitals used more hospital and ICU beds and more physicians per 1,000 decedents than the Sacramento benchmark, there was striking variation among the twenty-eight hospitals listed in Exhibit 4. Hospital bed inputs varied from 62.0 to 110.6 per 1,000; ICU beds, from 17.5 to 50.4 per 1,000; and FTE physician labor inputs, from 27.7 per 1,000 to 57.6 per 1,000.

Utilization of care. The contrasting patterns of care between the Sacramento benchmark and Los Angeles hospitals are quite striking. In managing chronic illness at the end of life, providers serving Los Angeles relied much more on inpatient care, aggressive use of ICUs and medical specialists, and frequent referrals, while care in the Sacramento region was characterized by greater reliance on primary care and parsimonious use of inpatient care, physician visits, and referrals. Yet there was considerable variation among Los Angeles hospitals in care intensity (Exhibit 5). During the last six months, patient days per decedent varied by a factor of 1.85 (12.7–23.5 days), and days spent in the ICU per decedent varied by a factor of 2.84 (4.0–11.4); the frequency of physician visits varied by a factor of 2.34 (39.7–92.8). Our measure of propensity for multiple referrals, the percentage of patients who saw ten or more physicians, varied by a factor of 2.06 (28.0–57.7 percent of decedents). Finally, the percentage of patients who died during a hospitalization that included an admission to the ICU (our measure of relative aggressiveness of terminal care) varied by a factor of 2.32 (21.6–50.0 percent of decedents).

Evaluating performance within California hospital systems. During the past decade or so, many U.S. hospitals have merged, been purchased, or otherwise become associated with multiple hospitals to form “hospital systems.” Many of these describe themselves as integrated health care systems. These hospital networks seem a logical place to seek accountability for resource allocation and for developing and implementing population-based approaches to managing chronic illness.

Among large systems. In California, we found much within-system variation. Appendix Figure 1 illustrates the variation among the three hospital systems with more than twenty hospitals in California.¹³ Medicare inpatient spending during the last two years of life varied by a factor of 2.17 among hospitals belonging to the Sutter Health system; by a factor of 2.72 within Catholic Healthcare West; and by a factor of 3.54 among Tenet hospitals located in California. The systemwide average for Sutter was $30,814; $29,802 for Catholic Healthcare West; and $46,323 for the Tenet hospital system. This figure shows that variation was also great for hospital days, days in intensive care, and physician visits per decedent.

Within the University of California hospital system. The hospitals in the University of California (UC) system enjoy strong reputations for high quality of care. For example, on the 2001 U.S. News and World Report list of hospitals with excellence in geriatric care, University of California, Los Angeles (UCLA), ranked first in the nation and University of California, San Francisco (UCSF), ranked eighteenth.¹⁴ Exhibit 7 provides estimates of Medicare spending and resource inputs during the last two years of life, by hospital. Medicare spending for inpatient care varied from $42,577 per decedent at University of California, San Diego (UCSD), to $57,721 at UCLA; Part B payments varied from $7,301 per decedent at UC Davis to $14,201 at UCLA. UC Davis used the fewest acute care beds (54.9 beds per 1,000); UCLA used the most (93.5). UCSF used the fewest ICU beds (12.2); UCLA used the most (50.4). Physician labor input varied both in per capita amounts and in the mix of primary and specialty care, reflecting the specialty-oriented practices at UCLA and UC Irvine and the primary care orientation at UCSF and UC Davis.

Exhibit 7 also compares utilization rates during the last six months of life. The most striking differences were those between UCSF and UCLA. UCLA, like many other hospitals in its region, managed chronic illness aggressively: Compared with UCSF, the population receiving care from UCLA spent 45 percent more days in acute care hospitals, used 3.49 times more days in the ICU, and was 1.53 times more likely to have an ICU stay during a terminal hospitalization. They experienced 1.71 times more physician visits and more frequent referrals (the percentage of decedents seeing ten or more physicians was 1.37 times greater at UCLA than at UCSF). The ratio of specialist to primary care visits at UCLA was 2.86; by contrast, UCSF’s ratio was 0.68. Although the two hospitals have similar scores on technical processes of care, UCSF is rated above average by its patients, while UCLA is rated average.

California hospitals and HRRs. Performance measures for California HRRs (including 95 percent confidence intervals) are available on the Dartmouth Atlas Web site. Inpatient measures are reported for hospitals with more than eighty deaths; physician measures are available for hospitals with at least 400 deaths.

Discussion

In this paper we used Medicare claims data to develop population-based, hospital-specific measures of Medicare spending, resource inputs, and use of care over fixed intervals of time for patients with severe chronic illnesses. We illustrated the importance of population-based measures by showing that per person spending during the last two years of life is determined more by the volume of services (patient days per decedent) than by the price per unit of service (average reimbursements per day in the hospital). We also demonstrated the importance of comparing performance at the hospital level. The hospital effect on utilization among California hospitals is ubiquitous across demographic factors, SES, category of illness, and severity of illness (as illustrated by the striking correlations between utilization rates for the last six months of life and rates nineteen to twenty-four months prior to death).

Our concept of evaluating relative efficiency is based on the notion of benchmarking: a comparison across regions and hospitals based on spending, resource inputs, and utilization measures and on available quality measures. In the example presented here, we first compared population-based measures in four regions, selecting Sacramento as the benchmark. We then used this benchmark to evaluate selected Los Angeles hospitals. Except for inpatient price, each hospital exceeded the Sacramento benchmark for every measure, even though there were wide variations among hospitals. Some might argue that evidence-based specification of the proper processes of care is required to identify efficient practices. But scientifically valid, detailed evidence defining efficient clinical pathways—for example, whom to hospitalize, when to schedule a revisit, or when to refer to a medical specialist, home health agency, or hospice—simply doesn’t exist. It will take a long time and a major reorientation of the academic research agenda to provide such clinical evidence, if indeed it is ever possible to do so. In the meantime, we argue that the results of natural experiments—population-based studies comparing overall quality and outcomes for similarly ill patients exposed to different levels of care intensity—should be used to establish benchmarks of relative efficiency. So far, these studies indicate no marginal gain from greater resource use across the range of practice observed within the United States. For this reason, we believe that the Sacramento region provides a fair benchmark for evaluating performance.

Our evaluation of performance involved three categories: relative spending, resource inputs, and utilization. Although some might prefer per person spending over fixed intervals of time as the gold standard, per person spending involves price, and price does not necessarily correspond to a hospital’s actual cost of producing care. Cost shifting between service lines and among payers, variations among hospitals in the proportion of patients with “outlier” payment status, and Medicare policies related to subsidies for indirect medical education (IME) and DSH payments distort price as an accurate summary measure for resource inputs per unit of care or per person over time.¹⁵ By contrast (in the absence of fraudulent billing), the claims-based measures of resource inputs—hospital beds, ICU beds, and FTE physicians per capita—estimate real differences in the amount of resources allocated to care for similarly ill patients among hospitals and across regions. Our measures of FTE physician by specialty also address a different aspect of the efficiency problem—namely, the number allocated and the mix among specialties. Our measures of utilization allow for the characterization of relative intensity of specific forms of care, including treatment of the terminally ill and the propensity to refer. Taken together, the measures provide a useful characterization of hospital-specific efficiency. For example, while per person spending for inpatient and Part B care during the last two years of life among patients receiving most of their care from UCLA’s hospital was only 1.26 times greater than for UCSF, UCLA used many more “real” resources in managing care: 1.66 times more physicians and 4.14 times more ICU beds per capita. Clinicians associated with UCLA also treated their chronically ill patients much more aggressively: 35 percent of deaths involved a stay in an ICU, compared with 23 percent at UCSF. The UCLA pattern of practice depended much more on medical specialty care than at UCSF, which emphasized primary care (Exhibit 7).

Limitations. Certain limitations of our measures need to be mentioned. Our utilization measures cover the last six months of life only. This is by design: We believe that variations during this period are not likely to be explained by differences in severity of illness. Moreover, the hospital-specific performance documented during this period of life is indicative of relative performance in managing chronic illness over longer periods of time, as shown by the high correlation with utilization rates observed in previous periods for the same patient cohort (Exhibit 2).

Certain limitations related to the availability of claims data should also be recognized: The measures of physician performance are based on a 20 percent sample of physician claims, and there is an eighteen-month time delay between date of use and availability of the claims for analysis. These limitations could be reduced by a change in CMS policy.

Finally, our measures do not address what is happening to younger patients. There is evidence that the variation in hospitalization rates seen for Medicare is highly correlated with variation for other chronically ill insured populations. There is also evidence that volume of care accounts for most of the regional variation in cost per capita among commercially insured populations.¹⁶ However, a hospital’s ranking in terms of per capita spending could vary greatly for commercial payers based on market-negotiated (rather than CMS-set) unit prices and the greater spending on nonchronic conditions such as pregnancy. We therefore cannot be confident that the association between per capita reimbursements and unit price of care seen in Medicare will accurately predict the relationships among other payers. The best strategy for addressing these limitations would be for all payers and self-insured employers to work together to produce resource input and utilization data for cohorts across Medicare, Medicaid, and commercially insured patients. The recently announced partnership between the California Public Employees’ Retirement System (CalPERS) and the Pacific Business Group on Health (PBGH) to build stakeholder consensus on a standard set of metrics for evaluating hospital efficiency of California hospitals is an encouraging development.

Policy responses. The availability of information on relative efficiency in managing chronic illness by specific providers could stimulate major employers and payers to use the data to direct their chronic disease populations away from high-cost, high-utilization hospitals to those that spend less and use fewer resources. The information should promote the profitability of plans participating in Medicare Advantage (MA) that can redirect their patients to physician groups using hospitals with spending levels below the regional reimbursement average that the CMS uses to calculate payment to plans. After commercial insurers and self-insured employers have substituted commercially negotiated unit prices into such analysis, changes in provider network composition or incentives for chronically ill patients to choose efficient providers should result in net savings for employers and payers. The potential savings in some markets are quite large. For example, during the study period, if the per person level of Medicare spending during the last two years of life for inpatient care and Part B in Los Angeles had been equal to the amount predicted by the Sacramento benchmark, Medicare would have saved $1.7 billion.¹⁷

Simply steering patients to low-cost providers, however, would result in little improvement in quality or system efficiency beyond that achieved by reducing overuse of supply-sensitive care among those who change providers. Ironically, unless traditional Medicare, too, can join in directing patients to efficient providers, it stands to lose; if MA plans, other commercial payers, and self-insured employers steer patients away from high-resource/high-volume providers, the populations loyal to high-cost providers will shrink, but the available resources will not, which will presumably result in still higher utilization rates and higher costs and possibly worse outcomes among chronically ill patients who remain loyal to such providers.

Chronically ill Americans need a fundamental redesign of care, shifting resources from the overused acute care sector to the now underfunded infrastructures of care for the management of patient populations. But achieving a major redesign requires new economic arrangements that pay for performance that actually (demonstrably) improves systemwide efficiency—that reward, rather than penalize, provider organizations that successfully reduce overreliance on acute hospital care and develop population-based strategies for managing their patients with chronic illnesses. Under current fee-for-service—and most forms of primary care capitation—the savings generated by effective population-based management of chronic care are returned to the payer, not the provider. Under these financial arrangements, providers that implement “best practice” models lose twice: Important aspects of the infrastructure of care go uncompensated, and reductions in acute inpatient care result in loss of revenue. What is needed is a financial plan that would share savings among payers and providers, after paying for the real costs of reducing capacity (for example, the loss of revenue targeted to amortize debt) and the cost of care under the population-based model for managing chronic illness.

The CMS is already moving to develop new strategies for promoting improved management of chronic illness. The Medicare Prescription Drug, Improvement, and Modernization Act (MMA) of 2003 directs the CMS to pay all hospitals based on resources needed for “efficient care.” We believe that our measures of spending, resource inputs, and utilization could be of use in pursuing this goal. Potentially the most flexible and innovative approach from the provider perspective is Medicare’s Health Care Quality Demonstration Programs (HCQDP), authorized under MMA Section 646. Regardless of the approach taken to address the overuse of care in managing chronic illness and the deficiencies in quality apparent in the delivery system, reducing excess acute care capacity carries some risks as well as benefits. The potential adverse consequences include loss of employment in health care, lowered hospital revenues that could affect both bond and equity markets, and worsening of access to care for the uninsured if safety-net hospitals are not protected.¹⁸ The consequences also could include worse health outcomes, if reductions in inpatient care are not associated with improvements that reduce the underuse of effective care and if there is no coordination and integration of care among other sectors involved in managing chronic illness—for example, home health and hospice care. Every community will have its own set of problems and its own potential for creative redesign. This is why flexibility in financing is so critical to genuine reform. These concerns also underscore the importance of ongoing performance monitoring of the health care system as change is implemented. Only then will we be able to learn what approaches to reform are most effective.

This study was supported by the Robert Wood Johnson Foundation, the California HealthCare Foundation, Aetna, the United Health Foundation, the WellPoint Foundation, and the National Institute on Aging (Grant no. P01-AG019783).

NOTES

1. J.E. Wennberg, E.S Fisher, and J.S. Skinner, “Geography and the Debate over Medicare Reform,” Health Affairs, 13 February 2002, content.healthaffairs.org/cgi/content/abstract/hlthaff.w2.96 (21 October 2005).
2. K. Baicker and A. Chandra, “Medicare Spending, the Physician Workforce, and Beneficiaries’ Quality of Care,” Health Affairs, 7 April 2004, content.healthaffairs.org/cgi/content/abstract/hlthaff.w4.184 (21 October 2005).
3. E.S. Fisher et al., “The Implications of Regional Variations in Medicare Spending, Part 1: The Content, Quality, and Accessibility of Care,” Annals of Internal Medicine 138, no. 4 (2003): 273–287; E.S. Fisher et al., “The Implications of Regional Variations in Medicare Spending, Part 2: Health Outcomes and Satisfaction with Care,” Annals of Internal Medicine 138, no. 4 (2003): 288–298; and J.S. Skinner, E.S. Fisher, and J.E. Wennberg, “The Efficiency of Medicare,” in Analyses in the Economics of Aging, ed. D.A. Wise (Chicago: University of Chicago Press, 2005), 129–157.
4. E.S. Fisher et al., “Variations in the Longitudinal Efficiency of Academic Medical Centers,” Health Affairs, 7 October 2004, content.healthaffairs.org/cgi/content/abstract/hlthaff.var.19 (21 October 2005).
5. J.E. Wennberg et al., “Use of Hospitals, Physician Visits, and Hospice Care during the Last Six Months of Life among Cohorts Loyal to Highly Respected Hospitals in the United States,” British Medical Journal 328, no. 7440 (2004): 607–610; and J.E. Wennberg et al., “Use of Medicare Claims Data to Monitor Provider-Specific Performance among Patients with Severe Chronic Illness,” Health Affairs, 7 October 2004,
content.healthaffairs.org/cgi/content/abstract/hlthaff.var.5 (21 October 2005).
6. See L.I. Iezzoni et al., “Chronic Conditions and Risk of In-Hospital Death,” Health Services Research 29, no. 4 (1994): 435–460.
7. Wennberg et al., “Use of Medicare Claims Data.”
8. The five performance measures for AMI are the percentage of eligible patients receiving (1) aspirin at time of admission; (2) aspirin at time of discharge; (3) angiotensin-converting enzyme (ACE) inhibitor for left ventricular dysfunction; (4) beta-blocker at admission; and (5) beta-blocker at discharge. The two CHF measures are the percentage of patients with (1) assessment of left ventricular function, and (2) ACE inhibitor for left ventricular dysfunction. For pneumonia, the three measures are percentage of patients with (1) oxygenation assessment; (2) pneumococcal vaccination; and (3) timing of initial antibiotic therapy. The summary scores are equally weighted averages for the items in each category. Hospital-specific summary scores are given only for those hospitals for which four of the five AMI and all of the CHF and pneumonia measures were based on twenty-five or more patients. See A.K. Jha et al., “Care in U.S. Hospitals—The Hospital Quality Alliance Program,” New England Journal of Medicine 353, no. 3 (2005): 265–274. (Regional scores in the current study are based on the average for each measure, obtained by summing numerator and denominator information across all reporting hospitals.)
9. California Institute for Health Systems Performance and California HealthCare Foundation, What Patients Think of California Hospitals: A Consumer Guide, 2004 ed. (California: CIHSP and CHCF, 2004).
10. CIHSP and CHCF, California Hospital Experience Survey, 2004 Edition Technical Summary, September 2004, www.chcf.org/documents/hospitals/CAHospitalExpSurvey04TechSum.pdf (21 October 2005). See also Jha et al., “Care in U.S. Hospitals.” Data are from the CMS Hospital Compare data set, www.hospitalcompare.hhs.gov.
11. For each hospital, we computed the implied price per day by dividing total reimbursements by total days spent in the hospital.
12. The differences in inpatient spending among the regions are not explained by differences in hourly wages for hospital workers. For example, the average hourly wage index (2003–2006) in Los Angeles was $31.08; in San Diego, Sacramento, and San Francisco it was $29.79, $32.09, and $38.98, respectively.
13. Appendix Figure 1 is available online at content.healthaffairs.org/cgi/content/full/hlthaff.w5.526/DC2.
14. “America’s Best Hospitals,” U.S. News and World Report (23 July 2001).
15. For example, among the five UC hospitals, outlier payments varied greatly as a percentage of total payments: from 8 percent of total at UCSD to 27 percent at UC Davis. Supplements for indirect medical education (IME) varied from 10 percent at UC Irvine to 20 percent at UCSF. The disproportionate-share hospital (DSH) supplement varied from 9 percent at UCLA to 16 percent at UC Irvine; net inpatient reimbursements per decedent during the last two years of life (total – DHS + IME) for 1999–2003 at UCSD were $30,600; at UCSF, $32,100; at UC Davis, $33,300; at UC Irvine, $38,000; and at UCLA, $42,700.
16. Among regions, the hospitalization rates for patients insured by Michigan Blue Cross Blue Shield are highly correlated with rates for Medicare fee-for-service (and both are highly correlated with hospital beds per 1,000). See J.E. Wennberg and D.E. Wennberg, eds., Dartmouth Atlas of Health Care in Michigan, 2000, www.dartmouthatlas.org/atlaslinks/michatlas.php (27 September 2005). A recent study by the U.S. Government Accountability Office (GAO) found that price contributed about one-third, and utilization about two-thirds, to variation in spending for hospital and physician services among regions. See GAO, Federal Employees Health Benefits Program: Competition and Other Factors Linked to Wide Variation in Health Care Prices, Pub. no. GAO-05-856, August 2005, www.gao.gov/cgi-bin/getrpt?GAO-05-856 (27 September 20005).
17. To estimate potential savings, we first multiplied the spending rate in Sacramento for inpatient and Part B services (Exhibit 3) by the number of deaths occurring in Los Angeles to predict total spending for Los Angeles if the Sacramento benchmark had applied; savings were then calculated by subtracting predicted spending from actual spending.
18. The evidence for overcapacity in acute-sector care resources, more so in some hospitals and communities than in others, has yet to be taken into account by Wall Street in evaluating hospitals’ financial prospects. The situation in Los Angeles is particularly interesting because of the requirement that existing hospitals be rebuilt or otherwise reconstructed to meet new standards for withstanding earthquakes. The recovery of investments to rebuild Los Angeles hospitals at their present level of capacity will depend importantly on Medicare’s willingness to continue to subsidize current inefficiencies in hospitals and regions with high resource use and high use rates. Recent CMS initiatives to promote efficiency in chronic disease management suggest a changing posture.

John Wennberg (john.wennberg{at}dartmouth.edu) is director of the Center for the Evaluative Clinical Sciences and the Peggy Y. Thomson Professor for the Evaluative Clinical Sciences at Dartmouth Medical School in Hanover, New Hampshire. Elliott Fisher is senior associate at the Veterans Affairs (VA) Outcomes Group in White River Junction, Vermont, and a professor of medicine and of community and family medicine at Dartmouth Medical School. Laurence Baker is an associate professor of health services research and policy at Stanford University School of Medicine, Stanford, California. Sandra Sharp and Kristen Bronner are research associates at Dartmouth Medical School.

Read related articles by Max Baucus, Thomas Priselac, Uwe Reinhardt, Leonard Schaeffer and Dana McMurtry, and Barry Straube.

DOI: 10.1377/hlthaff.w5.526
©2005 Project HOPE–The People-to-People Health Foundation, Inc.