A variety of approaches have been used to contain escalating hospital costs. One approach is intensifying price competition. The increase in price based competition, which changes the incentives hospitals face, coupled with the fact that consumers can more easily evaluate the quality of hotel services compared with the quality of clinical care, may lead hospitals to allocate more resources into hotel rather than clinical services.
To test this hypothesis we studied hospitals in California in 1982 and 1989, comparing resource allocations prior to and following selective contracting, a period during which the focus of competition changed from quality to price. We estimated the relationship between clinical outcomes, measured as risk-adjusted-mortality rates, and resources.
In 1989, higher competition was associated with lower clinical expenditures levels compared with 1982. The trend was stronger for non-profit hospitals. Lower clinical resource use was associated with worse risk adjusted mortality outcomes.
This study raises concerns that cost reductions may be associated with increased mortality.
The last two decades brought about fundamental changes in the organization and delivery of medical services in the United States as payers seek to control the escalation in health care expenditures. Policies addressing these issues have been of two types. The first relies on containing costs through control of the prices paid to providers, beginning with the Prospective Payment System (PPS) for hospitals in 1983, Resource Based Relative Value Units (RBRVUs) payment for physicians in 1992 and the most recently implemented prospective payment for nursing homes. Such prospective payment systems provide the same incentives to cut costs to all providers, irrespective of the markets in which they are located and the competitiveness of their markets. The other relies on changing the focus of competition among health care providers, from quality based competition  to price based competition .
These policy changes were successful in bringing about a deceleration in hospital revenues and expenditures growth [2,3]. Little is known, however, about what specific strategies hospitals adopted and the impact these strategies may have had on the quality of care patients receive. Previous studies  found that hospitals increased efficiency in all clinical services following selective contracting in California. California hospitals also tended to specialize and differentiate themselves from similar hospitals in response to competitive pressures . A similar response, of increased specialization, was observed for a national sample of hospitals following implementation of PPS .
In this paper we study another potential strategy that hospitals may adopt and which has not been addressed in the literature to-date. We investigate the hypothesis that hospitals in an increasingly price competitive environment, shift resources from activities related to clinical services, which are not easily observed and evaluated by patients, into hotel services which are easily observed. We study hospitals in California, comparing resource allocation during a regime dominated by quality competition and a regime dominated by price competition. We then examine the association between risk adjusted excess hospital mortality and resource use in clinical services, to investigate the potential impact on quality of medical care and health outcomes.
Competition in California in the 1980s: a case study
The implementation of selective contracting in California in 1982 offers a unique natural experiment to study the response of hospitals to changes in the nature of competition. Unlike other health care markets, in which the change from quality to price based competition was gradual, driven by continuously increasing penetration of managed care, and often time confounded by other secular trends, the California legislation changed market conditions very rapidly for all hospitals in the state by permitting all health plans for the first time to contract with only a subset of hospitals. It thus allows a pre/post study design: hospitals resource allocation decisions during the quality competition regime (pre period) can be compared to decisions made during the price competition regime (post period). This natural experiment allows us to test the hypothesis that changes in the nature of competition are more likely to be associated with a shift of resources from clinical to hotel activities (and a concomitant deterioration in mortality outcomes) in more competitive hospital markets. As the level of hospital competition has not changed during the period (the Herfindahl-Hirschman Index (HHI) remained stable in all markets), this test is limited to the change in the nature of competition and is not confounded by the impact of changes in the level of competition on resource allocation decisions.
During the same time period, California hospitals were also subject for the first time to price regulation, due to implementation of the Medicare PPS. While PPS also provided hospitals with incentives to lower their costs, it did so in a distinctively different manner than price competition. The PPS set the price per discharge hospitals were paid, thus creating incentives to lower costs, irrespective of market structure (7). The intensity of price based competition, on the other hand, is highly sensitive to the competitiveness of the hospital market. The analytical strategy of this paper is based on this distinction.
Hospital competition, quality, resource allocation and health outcomes
Competition focused on prices, as is often the case in markets dominated by managed care, creates incentives to increase efficiency and possibly curtail resource use. With the exception of possible increases in administrative activities designed to contain costs in other areas (e.g. billing and utilization review) or to increase marketing efforts, such incentives to cut costs are likely to affect all aspects of hospital activities.
Hospitals may also compete on quality, both quality of medical services and quality of hotel services and amenities. The importance of competition for quality is likely to be greater in markets in which hospitals compete for patients directly, as they do for all fee-for-service patients and for those enrolled in HMOs that offer a choice of hospitals within their market. Furthermore, to the extent that HMOs make their contracting decisions based on beneficiary hospital preferences, perceptions of quality are important competitive tools.
Competition for quality , unlike competition for price , may lead to increased costs. Furthermore, it may affect clinical and hotel services differently. In markets where patients' choice of hospitals are increasingly important, hospitals are likely to compete more on quality attributes that patients observe and value. Given the difficulty that patients have in directly determining the quality of medical care they receive, and the relative ease with which they can evaluate the quality of hotel services (e.g. condition of the facility, quality of food) hospitals face incentives to shift resources from clinical activities to amenities. On the other hand, if patients rely on their physician's recommendations in choosing hospitals , and to the degree that physicians can assess clinical quality, albeit imperfectly, hospitals are faced with counter incentives, incentives that would promote resource use in clinical activities rather than hotel services. As a result, hospitals may face conflicting incentives: incentives to maintain or enhance the quality of hotel services on the one hand, and incentives to maintain activities that contribute to the quality of clinical care and health outcomes on the other. The actual choices that hospitals make about resource allocation depend on the relative strength of these opposing incentives.
A model of changes in resource allocation
As the main hypothesis of interest is that the change in the nature of competition was associated with changes in hospital resource allocations, the model we hypothesize allows the marginal effects of market and hospital characteristics (Xt) on resource allocation (Yt), as measured by the coefficients in a regression model (βt), to vary over time:
(1) Yt = αt + βtXt + εt
The models we estimated were difference models of the form
(2) ΔY = Δα + Δβ * X0 + βt * ΔX + Δε
where Δ is the difference between year t and the base year, indicated by t = 0. From (2) it follows that βt, the vector of coefficients multiplying the change variables, measures the marginal effect of the variable in the end year, while Δβ, the vector of coefficients multiplying the level variables, measures the change in the marginal effect. The marginal relationship in year 0 is given by β0 = βt - Δβ..
The initial sample included all 338 acute care hospitals in California that were in operation during both 1982 and 1989. Of those, 18 (5.3%) were excluded from the resource allocation analyses and 8 (2.4%) were excluded from the mortality analyses, because of incomplete data.
Financial, ownership and utilization data were obtained from the Hospital Annual Financial Disclosure Reports, filed annually by all California hospitals with the Office of Statewide Health Planning and Development (OSHPD). Risk adjusted mortality data were obtained from the Medicare Hospital Information Report published by the Health Care Financing Administration .
I. Resource allocation variables
Resource allocation was measured by expenditures per adjusted discharge. Adjusted discharges are a composite measure of input designed to account for both inpatient discharges and outpatient visits, using the methodology developed by the American Hospital Association.
Expenditures per adjusted discharge were calculated separately for three categories: clinical, hotel and administrative services. Expenditures were aggregated by cost center, with each cost center assigned to one of the three services. The table in the 1 lists all hospital cost centers and their assignment to hotel, clinical and administrative categories. The dependent variables in the resource allocation analyses were defined as the differences in expenditures per adjusted discharge, between 1989 and 1982, for each of the three categories.
II. Quality variables
The dependent variables for the analyses of quality of clinical care were excess death rates from all causes and from 4 specific medical conditions that have relatively high death rates: acute myocardial infarction, congestive heart failure, pneumonia and stroke. We included in the analyses measures based on cause specific mortality in addition to overall mortality because prior studies  have shown that these measures tend to be uncorrelated, and that hospitals performing well in one clinical area do not necessarily perform well in others. Measures based on overall mortality may therefore be biased towards zero, showing less variation compared with cause specific measures.
Excess mortality was defined as the difference between the observed mortality rate for the hospital and a predicted, risk adjusted mortality rate. Observed and predicted mortality rates were obtained from the Medicare Hospital Reports . They are based on Medicare discharges and include all deaths within 30 days of admission, irrespective of the location of death. The risk adjustment methodology used by the Health Care Financing Administration, incorporates individual patients' age, gender, specific diagnoses and comorbidities, admission source, emergency or elective admission and the patient's risk group based on hospitalizations during the preceding 6 months .
III. Independent variables
Competition was measured by the Herfindahl-Hirschman Index (HHI), defined as the sum of squared market shares of all hospitals competing in the same area. Hospital market areas and the HHI were calculated based on all payer zip code level patient flows, as described in Zwanziger et al. .
To control for financial pressures hospitals may have been experiencing in addition to competition, we included variables measuring bad debt and charity as percent of total revenues and percent occupancy. To control for potential economies of scale the estimated models included total clinical standard units of measures reported in the California Financial Disclosure Reports. Ownership indicator variables included for-profit, not-for profit public, and not-for profit district ownership. The omitted category was private not-for profit hospitals. Teaching status was defined as hospitals with some residents. All payer DRG-based case mix index was included to account for differences in patients' severity. Median family income measured for the hospital's zip code area was included to capture demand effects and as a proxy for cross sectional wage variations.
We estimated regression models in which change in expenditures per adjusted discharge and excess mortality in 1989 were the dependent variables. Because all models were heteroskedastic, all reported tests of significance are based on White's robust standard errors . The Ramsey RESET test for specification errors  was applied to all models to rule out the need for non-linear and interaction terms.
The mortality models were weighted by the inverse of the standard error for the predicted mortality rate, to account for differences across hospitals in the accuracy of the excess mortality measures, which are due to differences in sample sizes .
Since initial analyses indicated different associations (different βs) for for-profit and non-profit hospitals, we estimated fully interacted models, in which all variables were interacted with for-profit status. The hypotheses of significant marginal effect were therefore tested for the non-profit hospitals by a t test of the main effect and for the for-profit hospital by an F test of the linear restriction that the sum of the coefficients of the main and interaction effect are zero.
Description of sample hospitals
Table 1 presents descriptive statistics for the hospitals included in the study. The majority of hospitals (52.6%) were private non-profit with the second largest group (26.0%) being for-profit institutions. Fifteen point four percent were teaching hospitals. The average hospital size did not change significantly over the 1982–1989 period, remaining at 190–200 beds. Occupancies declined significantly, from an average of 62.3% to 55.2%, and inpatient case mix increased significantly from 1.17 to 1.27, indicating that hospitals were treating sicker and more expensive patients at the end of the period. Both total expenditures and expenditures per adjusted discharge increased significantly.
Table 1. Descriptive Statistics
The degree of competition among hospitals has not changed between 1982 and 1989. The HHI of around 0.3 suggests that competition was limited. (Markets with HHI values below 0.18 are considered moderately or very competitive ). The large variation in the HHI, however, indicates that many hospitals were located in competitive markets, with 25% of hospitals in markets with HHI below 0.17.
Overall mortality rates averaged 10%. Average observed and predicted rates were very similar, but the variation in rates was higher for the observed rates compared with the predicted rates, suggesting substantial variations in excess mortality and quality across the sample.
Resource allocation changes
Table 2 reports the mean values for the dependent and independent variables included in the multivariate regressions. Table 3 reports results by ownership – for-profit and non-profit. These results are based on a fully interacted model estimated over pooled data by ownership. All models were highly significant (p < 0.01). The clinical services model explained 51% of the variation in expenditures per adjusted discharge, while the hotel and administrative services models explained 24% and 26% respectively.
Table 2. Means and standard deviations of variables included in the multivariate analyses
Table 3. Expenditures per adjusted Discharge – Multivariate regression results
Effect of competition
Table 5 presents the marginal effect of competition, calculated from the regression results and using equations 2 and 3, for 1982 and 1989, as well as the change in these coefficients between the two years. (Note that the regression coefficients for the HHI variable were multiplied by -1 in table 5, such that a positive association means that expenditures per discharge increase with increased competition.)
Table 5. Marginal effect of competition on Expenditures per adjusted discharge (Positive values indicate increase with increased competition)
Non-profit hospitals in more competitive areas had higher expenditures per adjusted discharge in all three categories in all years. The marginal effect was highest in clinical areas and lowest in administrative services. It declined significantly over time in both clinical and hotel services, but not in administrative services. The decline was almost three times as large in the clinical services compared with hotel services. By 1989, while the marginal effect of competition on expenditures in these services was still positive, it was no longer significantly different from zero. These findings are consistent with the hypothesis that the focus of competition on quality in 1982 has diminished significantly over the seven-year period we studied.
The results for for-profit hospitals present a different picture. First, the association between expenditures per adjusted discharge and competition was statically significant only for administrative services. This association was also by far the strongest. It was negative, indicating that hospitals in more competitive markets spent less per discharge on administrative activities. The marginal effect was slightly smaller in 1989, suggesting that hospitals in more competitive areas may have reallocated resources into administrative services. They may have, for example, invested in better information and management systems that would allow them to better control costs.
The association between competition and resource use was negative in clinical services and positive in hotel services. This is consistent with the hypothesis that for-profit hospitals compete on quality in those areas that can be easily observed by patients, namely hotel services, and cut back on resources in clinical services, where quality is more difficult for patients to evaluate directly. The change over the 1982 through 1989 period is also consistent with this hypothesis: the negative association between competition and clinical resources increased in 1989 as did the positive association between hotel resources and competition. The lack of significance of the associations may reflect the smaller number of for-profit hospitals in the sample and the resulting lower statistical power. (There were 83 for-profit hospitals and 13 independent variables, compared with 237 non-profit hospitals.)
Other hospital and market characteristics
The strongest and most consistent relationship was between all payer case mix and expenditures per adjusted discharge in all categories (see table 3). The association was substantially stronger for the clinical category and in non-profit compared with for-profit hospitals.
Most other variables either exhibited no significant associations or no clear patterns. There were no significant differences in resource allocations by ownership. Teaching status was positively associated with clinical and hotel expenditures among the non-profit hospitals but exhibited a negative association among the for-profit hospitals. Percent bad debt and charity and percent occupancy had no significant relationship with expenditures among the non-profit but were associated with lower clinical and administrative expenditures among the for-profit hospitals. Median family income was associated with higher expenditure levels for all services among the non-profit hospital, but only with the clinical services among the for-profits.
Risk adjusted excess mortality
Table 4 reports the results of the regressions modeling the association between excess mortality (defined as the difference between observed and predicted mortality rate) and clinical expenditures per adjusted discharge, competition, ownership, and teaching status. The models explained between 1% and 5% of the variation in excess mortality. In all cases there was a negative association between clinical expenditure levels and excess mortality, implying that increased resources were associated with better mortality outcomes. This relationship was present for mortality from all causes as well as from the four specific causes, and was statistically significant at the 0.10 level or better.
Table 4. Risk adjusted excess mortality – Multivariate Regression Results
Table 6 shows the increase in excess mortality that is associated with a decrease of 1 standard deviation (SD) in clinical expenditures per adjusted discharge, based on the estimated regression coefficients. For comparison, the table also provides the magnitude of 1 SD in excess mortality among the study hospitals. In all cases, the effect of a 1 SD in resources was less than a 1 SD in excess mortality. As expected due to the potential bias towards zero in the measures based on all causes, the associations were larger for the cause specific measures.
Table 6. Increase in excess mortality rates associated with 1 standard deviation decrease in clinical expenditures per adjusted discharge
As discussed earlier, we hypothesized that competition may affect quality, including clinical quality, not only through its impact on resource use but also due to incentives to compete on quality. If indeed hospitals were competing on clinical quality, the association between the HHI and excess mortality, controlling for resource use, should have been positive and significant. In all cases, except for pneumonia, we did not find a significant relationship. In most cases there was also no significant relationship between ownership or teaching status and excess mortality.
In this paper we present a test of the hypothesis that changes in the nature of competition among California hospitals, resulting from selective contracting, were associated with changes in hospitals' resource allocation decisions. We find empirical evidence to suggest that resources have been shifted from clinical activities (which are not observed by patients) and into hotel services (which are more readily observable). These changes in resource allocation tended to be larger in hospitals located in more competitive areas. As the level of competition has not changed during the study period, the change in hospital behavior is likely to be a response to the change in the nature of competition.
The relationship between resource allocation and competition differed by ownership. For-profit hospitals in more competitive areas had lower expenditures levels compared with those in less competitive areas. Among non-profit hospitals, we found the opposite – clinical expenditures per adjusted discharge increased with competition. The trend over time, however, even though it was much stronger among the non-profit hospitals, was the same for both types. For both, the change in the marginal effect of competition (Δβ) was negative. As a result, the positive association between competition and clinical resource use among non-profit hospitals diminished and the negative association among the for-profit hospitals increased.
The analyses of excess mortality demonstrate that clinical quality, at least as measured here, is positively associated with the amount of resources used in producing clinical services. Therefore, policies that create incentives for hospitals to limit resource use are likely to have an impact on health outcomes. Furthermore, as clinical quality is not easily observable by consumers, leading to disparity in incentives to provide hotel and clinical quality, more competitive areas are likely to experience a larger relative decline in resources allocated to clinical activities and hence worse mortality outcomes.
The impact on health outcomes in non-profit hospitals may not be as large as it might have been because expenditure levels were curtailed not only in clinical areas, but in hotel services as well, although to a much lesser degree. This strategy spread the burden of cost containment efforts beyond clinical activities. If non-profit hospitals would have concentrated all their cost cutting efforts in clinical services, as did the for-profit hospitals, the impact on costs, and potentially on mortality outcomes, would have been 35% higher (see table 4).
The generalizability of the findings presented here is limited in several ways. First, quality was measured only in terms of excess mortality. While this is an important aspect of quality, it is likely to be an insensitive measure. Because measures based on mortality do not tend to be correlated with measures based on other outcomes , such as complications, one cannot deduce from this study that other aspects of clinical quality have been affected similarly by the changes in competition during the period. Furthermore, the mortality models we estimated, unlike the expenditures models, were cross sectional and are therefore subject to the usual concerns about potential bias due to omitted hospital specific effects. It should be noted, however, that the models did include the variables most likely to be associated with excess mortality – patient level risks, expenditures, competition, ownership and teaching status. It should also be noted that while our findings with respect to changes in resource allocation are based on total expenditures, thus reflecting care for all patients, the mortality outcomes are based on the experience of Medicare patients only. Prior studies, however, suggest that patient care given by same provider does not vary by payer status [17-19].
A more important generalizability question arises due to the sample selected for this study – namely hospitals located only in California during the 1980s. Can the behavior of hospitals observed in this local and during this time period be assumed to generalize to other markets and other times? While the magnitude of the effects we measure in this study are clearly not generalizable, the directions of the effects are likely to transcend time and place, as they reflect basic behavioral responses to market incentives in accordance with theory. While this sample selection may be viewed as a limitation of this study, it is also its strength. By focusing on a period in which levels of competition were stable and where the change in the nature of competition can be traced to a specific legislative act, this study is able to test for the differential impact of price vs. quality competition, without confounding by other factors.
DM designed the study, performed the analyses, and wrote the manuscript. AB prepared the data set, and JZ participated in the design and analyses.
The authors gratefully acknowledge funding from the Agency for Health Care Policy and Research (grant No. HS09545).
American Journal of Public Health 1995, 85:1391-1396. PubMed Abstract
Medical Care 1994, 32:1153-1162. PubMed Abstract
Health Services Research 1990, 25:757-83. PubMed Abstract
Health Care Financing Review Annual Supplement 1991, 45-77. PubMed Abstract
American Journal of Public Health 1997, 87:429-433. PubMed Abstract
New England Journal of Medicine 1989, 321:1720-1725. PubMed Abstract
Health Services Research 1994, 29:75-93. PubMed Abstract
Health Services Research 2000, 35:849-867. PubMed Abstract
The pre-publication history for this paper can be accessed here: