Health care issues
Volume 27, Issue 2, March 2008, Pages 339–361
234972||
Medicare reimbursement, nurse staffing, and patient outcomes
Robert Kaestner a, b, , ,
a Department of Economics and Institute of Government and Public Affairs, University of Illinois at Chicago, United States
b National Bureau of Economic Research, United States
c American Medical Association, United States
Received 20 March 2006, Revised 25 April 2007, Accepted 30 April 2007, Available online 29 November 2007
Show more Show less
doi:10.1016/j.jhealeco.2007.04.003
Abstract
There is widespread concern about the quality of health care in the US, and the effect of provider payments on the quality of care is an important and unsettled issue in this debate. The critical question is whether changes in provider payments affect health. To date there is relatively little research on this question. Here, we present evidence of the effect of plausibly exogenous changes in Medicare reimbursement – caused by geographical reclassification – on hospital staffing (nurses) and patient outcomes. We find that changes in Medicare reimbursement levels of approximately 10% have no meaningful effect on hospital use of resources or patient outcomes.
JEL classification
I12;
Keywords
Medicare;
Quality;
Nurses;
Mortality
1. Introduction
The quality of health care is an issue of significant concern for our society. Many Americans believe that the quality of health care is poor and that the quality of care has decreased over the last several years (Employee Benefits Research Institute, 2004 and Kaiser Family Foundation, 2004a). Moreover, concern about the quality of health care is not limited to the lay public. The Institute of Medicine (IOM) issued a series of influential reports on the quality of health care in the United States that portrayed a grim picture: “Quality problems are everywhere, affecting many patients. Between the health care we have and the care we could have lies not just a gap, but a chasm (Committee on Quality of Health Care in America, 2001, p. 1).” Indeed, the IOM (Kohn et al., 2000) claim that medical errors result in 98,000 unnecessary deaths per year in the United States has received considerable media attention and it has become a rallying cry for professional organizations such as the Leapfrog Group whose mission is to improve the quality of health care.1
Public opinion about the quality of health care is closely tied to views of managed care.2 The public is concerned that providers may respond to supply side incentives by limiting utilization and providing too little care, or what Newhouse (2002) and others refers to as “stinting” (Newhouse, 2002 and Ellis and McGuire, 1986). Supply side factors, in particular inadequate provider payments, are a cause of particular concern about the quality of health care in our public systems. Advocates for the poor worry about the quality of care received by Medicaid recipients because of inadequate provider payments.3 Medicaid payments are notoriously low and state policymakers continue to use payments to medical providers more as a fiscal tool than a health policy tool (Zuckerman et al., 2004; Lewin Group, 2006).4 Similarly, Medicare recipients and advocates for the elderly are concerned about the quality of care because of the federal government's use of provider payments as a way to control Medicare costs, as exemplified by the Balanced Budget Act of 1997 (Guterman, 1998 and Medpac, 2000).
The fear that inadequate payments to medical providers are adversely affecting the quality of health care is well founded. There is abundant evidence that providers respond to financial incentives by altering their treatment practices or case mix.5 So it is quite plausible that limiting or reducing provider payments may alter treatment, for example, reduce the amount of care or change the types of care received, and as a result, may adversely affect health outcomes. Surprisingly, empirical evidence to support this hypothesis is scarce. For example, two recent reviews by Miller and Luft, 1997 and Miller and Luft, 2002 conclude that managed care (versus fee-for-service) does not adversely affect health even though there is evidence that managed care was effective at reducing payments to providers and reducing utilization (Glied, 2000, Cutler et al., 2000 and Polsky and Nicholson, 2004).
In this paper, we study the effect of changes in Medicare reimbursement caused by geographical reclassification on hospital use of resources (e.g., nurses) and patient outcomes. Geographical reclassification of hospitals is an explicit policy response to inadequate provider reimbursement. Hospitals with relatively high labor costs receive higher Medicare reimbursement to ensure that they are financially sound and to ensure Medicare recipients have access to quality care (Scanlon, 2002). While previous studies have shown that providers alter treatment practices in response to changes in Medicare reimbursement, relatively few studies have examined the link between Medicare payments and health outcomes.6 However, information on this issue is a necessary part of a full assessment of the welfare implications of changes in Medicare payments (Newhouse, 2002). For example, reduced payments may eliminate care that has little benefit and for which the costs are greater than the benefits. In this case, reduced payments may be welfare improving. Alternatively, lower payments may eliminate more important types of care and adversely affect health. Therefore, our primary purpose in this paper is to add to the literature on the effect of the Medicare reimbursement on health.
Several previous studies have investigated this issue, mostly in the context of the switch to Prospective Payment System (PPS) in the early 1980s. Evidence from these studies is mixed, although it is clear that the switch to a PPS decreased hospital length of stay, and in the case of nursing homes decreased nursing resources (Hodgkin and McGuire, 1994, Cutler, 1995, Ellis and McGuire, 1996, Grabowski, 2001 and Konetzka et al., 2004).7 In a series of influential papers, Rogers and colleagues (Rogers et al., 1990, Kahn et al., 1990 and Kosecoff et al., 1990) examined the effect of PPS on patient health (adjusted for health at admission) using a nationally representative sample of Medicare patients pre- and post-PPS. Findings from these studies show no increase in mortality (in-hospital, 30-day, and 180-day) subsequent to the switch to the PPS, although the authors do report an increase in the number of patients discharged in an unstable condition (Kosecoff et al., 1990).8 Cutler (1995) also finds that the introduction of the PPS had no long-run effect on mortality of the elderly treated for severe illnesses. Cutler (1995) finds that decreases in average payments compress the mortality distribution: increase in-hospital mortality, decrease post-discharge mortality, no change in 1-year mortality.9
Two studies closest to ours are Staiger and Gaumer (1995) and Shen (2003). Both studies use plausibly exogenous variation in Medicare reimbursement rates in the post-PPS period to examine the effect of reimbursement on mortality of Medicare patients treated for urgent care and heart attacks. Also like us, both studies include hospital fixed effects and examine changes in mortality within hospitals pre- and post-changes in Medicare reimbursement. Estimates in Staiger and Gaumer (1995) suggest that changes in reimbursement have mixed and unintuitive effects. Reductions in Medicare payments increase mortality (45 days), but mostly for government hospitals, and to a lesser extent for-profit hospitals. In the case of not-for-profit hospitals, reductions in reimbursement are not significantly related to mortality. For all types of hospitals, Staiger and Gaumer (1995) find that reductions in Medicare reimbursement have no statistically significant effects on 1-year mortality. Shen (2003) limits the analysis to Medicare patients treated for acute myocardial infarction (AMI). She finds that reductions in Medicare reimbursement increase short-term mortality, but leaves 1-year mortality basically unchanged.
This brief literature review leads to a few conclusions. First, for such an important issue, there is relatively little research examining the effect of Medicare reimbursement on health and much of it pertains to a period when Medicare reimbursement was relatively generous. Second, results from this literature are mixed, although most studies find that reductions in Medicare reimbursement rates compress the mortality distribution. Third, the range of outcomes examined have been quite limited and basically restricted to mortality and readmission. Fourth, no study has used a research design that uses hospitals in close geographic proximity (e.g., within state) as a comparison group. This may be important given that there is a great deal of geographic variation in hospital treatment practices and health outcomes, and even studies that use hospital fixed effects methods such as Staiger and Gaumer (1995), are unable to control for time-varying, area-specific effects. Finally, no prior study has examined whether changes in Medicare reimbursement has spillover effects on the health of non-Medicare patients, for example, as hypothesized in Glazer and McGuire (2002). In this paper, we add to the evidence of the effect of changes in Medicare reimbursement on health using a plausibly exogenous change in Medicare reimbursement during a period when reimbursement rates were relatively low. We extend the literature by examining a broader range of outcomes; by using geographically close hospitals as controls; and by explicitly investigating whether there are spillover effects for non-Medicare patients.
2. Theoretical considerations
Our interest is to assess the effect of geographical reclassification on hospitals’ use of resources and patient outcomes. This is equivalent to examining the effect of an increase in Medicare reimbursement because the purpose of reclassification is to raise payments to hospitals. Chalkley and Malcomson (2000) present a simple model similar to others used in the literature that can be used for this purpose (e.g., Ellis and McGuire, 1986 and Hodgkin and McGuire, 1994). In this model, reimbursement is prospective, as in Medicare, and does not reflect actual resource use for each patient. Further, this model assumes that demand responds to changes in quality. The only way to increase the number of patients is to increase quality. Given this set up, an increase in reimbursement will raise quality. We measure quality by patient outcomes. To increase quality, the hospital will have to increase the use of resources that produce quality, for example, nurses.
This simple intuition motivates our empirical analysis, which will examine the effect of an increase in Medicare reimbursement caused by geographical reclassification on use of nurses, quantity of patients (admissions and inpatient days), and patient outcomes. We focus on nurses because they represent the largest resource in terms of cost used by hospitals and therefore it is likely that changes in quality would occur partly from a change in the quantity (or quality) of nurses. Note also, that the change in Medicare reimbursement caused by geographical classification does not alter the relative profitability of different Medicare patients because all DRGs experience the same proportionate change in reimbursement. Consequently, geographical reclassification will not create an incentive to reclassify patients (upcode), and there should be no compositional change pre- and post-reclassification in patient severity within DRG and no change in the degree to which payment is prospective (McClellan, 1997).
One issue related to changes in Medicare reimbursement is whether there will be spillover effects. For example, Glazer and McGuire (2002) assume that there is a common level of quality across patient types. In this case, any increase in quality as a result of an increase in Medicare reimbursement will also increase the quality of non-Medicare patients. Similar assumptions about common quality are found in other contexts and empirical evidence generally supports the hypothesis of common quality (Grabowski et al., 2006; Gertler and Waldman, 1992; Dranove and White, 1998).
However, there is also considerable evidence that the quality of patient care in hospitals depends on payer type, contradicting at least the extreme form of common quality. For example, Doyle (2005) found that trauma patients received different amounts of care, as measured by length of stay, number of procedures, and total charges, depending on payer status, and that these differences in care resulted in differential mortality. In this case, quality was different because of different lengths of stay and because of different intensity of treatment (number of procedures). Interestingly, the different number of procedures and different intensity of treatment would require different amount of nursing resources. Thus, there would be a different number of nurses (or other resources) assigned to a floor (unit) depending on the payer mix of the patients. Meltzer et al. (2002) also found differences in hospital resource use, as measured by charges, between Medicare and non-Medicare (private) patients in a specific hospital with the same diagnosis (DRG). Finally, the dramatic rise in the number of physicians who only treat patients in the hospital (hospitalist) and who often work for the hospital provides an institutional mechanism for hospitals to deliver differential amounts of care to patients of different payer types. Several studies have documented that hospitalists reduce length of stay and resource use of patients (Meltzer et al., 2002; Diamond et al., 1998; Rifkin et al., 2004). In short, there are plausible mechanisms for nursing and other resources to be allocated on a patient-type basis, and thus it is an empirical question whether an increase in Medicare reimbursement will have spillover effects on non-Medicare patients.
3. Research design and methods
Our objective is to obtain estimates of the effect of changes in Medicare reimbursement on hospital use of nursing and other hospital resources (e.g., beds) and patient outcomes for both Medicare and non-Medicare patients. While Medicare reimbursement levels change from year-to-year, these changes are mostly determined by national and regional factors that do not vary significantly within MSA (region) over time. For example, all hospitals in the same MSA will receive the same year-to-year increases in Medicare reimbursement due to local variation in labor costs, which do not exhibit much geographic variation. This component is referred to as the operating wage index. Moreover, much of the remaining variation in year-to-year increases in Medicare reimbursement has a national component (changes in DRG weight) that is common to all hospitals in the country. Analyses that hold constant differences in outcomes due to geography and time would leave little variation in Medicare reimbursement that could be exploited to identify effects of reimbursement on hospital resource use and patient outcomes. Previous studies have addressed this issue by using the variation in reimbursement brought about by introduction of the PPS system, but since this introduction, there has been little variation in Medicare reimbursement for inpatient care that is independent of year and geographic fixed effects.
3.1. Geographic reclassification
One source of variation in Medicare reimbursement is geographic reclassification. Hospitals can be reclassified to areas that receive higher Medicare reimbursement (operating wage index and standard payment). To qualify for such reclassification, a hospital has to meet two criteria.10 First, the hospital has to be within a specified number of miles of the target area: 15 miles for a metropolitan hospital and 35 miles for a non-metropolitan hospital. The hospital may still qualify for reclassification if it does not meet the proximity standard if at least half of the hospital's employees reside in the target area. Second, the metropolitan (non-metropolitan) hospitals’ average wages have to be 8% (6%) higher than the average wages in its assigned area and its wages have to be at least 84% (82%) of the target area's average wage.11 Rural and sole community hospitals can be reclassified by meeting less stringent criteria. Thus, hospitals that are most likely to be reclassified are those that are near, but not in, more densely populated and therefore higher wage areas, and rural hospitals. Hospitals apply for reclassification at least 1 year before reclassification will take effect, and they know the outcome of the reclassification application at least 6 months in advance of it taking effect. Prior to 2002, reclassification was valid for 1 year and hospitals had to apply every year to maintain the reclassification. Many hospitals that obtain reclassification subsequently lose their status because they cannot meet the wage requirements. These hospitals, which we refer to as declassified hospitals, also experience significant changes in reimbursement, which we exploit.
The geographic reclassification process has a few empirical implications. First, since hospitals know at least 6 months in advance that they will be receiving an increase in Medicare reimbursement, they may begin to respond immediately—prior to reclassification becoming effective. The empirical analysis should take this into account, which we do by examining changes between year (t − 2) and (t). Second, many reclassifications are temporary and hospitals’ responses may be smaller than if the change was permanent. Among hospitals that changed geographic area, 50% were reclassified for only 1 or 2 years and 50% were reclassified for three or more years. We examine whether estimates differ by the length of time reclassified. Third, and perhaps most importantly, is the question of whether geographic reclassification is exogenous; conditional on observed characteristics, is reclassification unrelated to hospital resource use and patient outcomes?
To qualify for reclassification, the hospital has to have unusually high labor costs for its area. Determination of labor costs is made by Centers for Medicare and Medicaid using data from 3 years prior for all hospitals in a reimbursement area (e.g., MSA or rural part of a state). So, it is very difficult for a hospital to manipulate costs, 3 years prior to a decision, to obtain reclassification in the current period. Rather, it is more likely that unusually high labor costs are purely a result of geography and/or hospital characteristics. If the cause of high labor costs is purely geography, which is consistent with the intent of the reclassification regulation, conditioning on geography and year may be sufficient to plausibly establish exogeneity. However, it would require conditioning on a geographic level that is difficult to identify—hospitals in a given proximity to a Medicare reimbursement boundary. And more importantly, there may be few hospitals meeting such a definition in any given year.
It is also likely that high labor costs may be due to systematic differences in hospital preferences or characteristics. For example, a hospital that has a strong preference for quality may employ higher quality employees that are more expensive. In this case, it is necessary to control for hospital specific effects to bolster the case for exogeneity. Hospital fixed effects account for any time-invariant characteristics of the hospital that are correlated with reclassification status, and staffing and patient outcomes. Thus, if unmeasured time-invariant factors are the source of endogeneity, the fixed effects procedure is effective at controlling for this problem. Given that most hospitals enter the data for only one experimental period (pre- and post-classification), most characteristics of the hospital will be relatively stable and the identifying assumption underlying the fixed effects specification gains in plausibility.
3.2. Empirical specification
Our research design, which is based on a multivariate regression, is intended to support the assumption that changes in Medicare reimbursement due to geographic reclassification are exogenous. Accordingly, we selected the sample and specified the regression model in ways consistent with this objective. First, we chose a sample that consists of a group of hospitals that were reclassified, or that lost reclassification status (declassified), in year (t) and a group of hospitals in the same geographic area and year that did not change status. We refer to hospitals changing status as treatments and the hospitals not changing status as controls. For each hospital, treatments and controls, we include 2 years of information: one from year (t − 2), which is before reclassification, and one from year (t), which is the year reclassification becomes effective. Then, as noted above, to bolster the exogeneity assumption, we control for either geographic area-year fixed effects, or hospital fixed effects and year.
We obtain estimates of the effect of changes in Medicare reimbursement on hospital use of nursing and other hospital resources (beds, admissions, inpatient days), and outcomes for both Medicare and non-Medicare patients. To make things concrete, consider the following regression model that relates the number of registered nurses (RN), which is measured at the hospital level (i), to whether a hospital was reclassified.
equation(1)
RNijt=β0+β1 TREATi+β2(TREATi xPOSTt)+XitΓ+δjt+uijt, i=1,…,N (index of hospitals), j=1,…,M (index of areas) t=1994,…,2001 (index of years)RNijt=β0+β1 TREATi+β2(TREATi xPOSTt)+XitΓ+δjt+uijt, i=1,…,N (index of hospitals), j=1,…,M (index of areas) t=1994,…,2001 (index of years)
In Eq. (1), the variable TREAT is an indicator (equal to one and zero otherwise) of whether a hospital was reclassified; POST is an indicator of whether the observation is post-reclassification; the vector (X) represents measured characteristics of the hospital such as whether it was a teaching hospital and whether it is a for-profit hospital. In analyses of the effect of reimbursement on hospital use of nursing resources, X will also include the number of beds (hospital and nursing home units), number of inpatient admissions (hospital and nursing home units), and number of outpatient visits. We include these “quantities” because our interest is in per-patient resources. We are also interested in the effect of reimbursement on quantity (e.g., inpatient days) and in these analyses we do not include these “quantity” variables. Finally, in analyses we present below, we replace the variable TREAT with two dummy variables indicating that the hospital was reclassified or declassified.
Eq. (1) also includes controls for geographic area-year effects (δjt). The geographic unit is the area used by Medicare to establish different reimbursement rates, and these are primarily urban areas (MSAs) in a state with all non-urban areas (i.e., rural areas) combined into one geographic unit. We allow each geographic area to have a separate effect in each year (eliminating the main effect of POSTt). For example, assume that in 1998 a hospital outside of Chicago was reclassified into the Chicago metropolitan area for Medicare reimbursement purposes. This hospital would contribute two observations: one in 1996 and one in 1998. We would also include other hospitals in the originating (or alternatively the destination) geographic area and each of these would contribute 2 years of information: 1996 and 1998.
In Eq. (1), we control for all geographic area-year combinations such as that described for this Chicago-area hospital. We are comparing treatment hospitals to control hospitals in the same area and year. Thus, there would be very little independent variation in the operating wage index unless some hospitals were reclassified. The average change in the operating wage index following reclassification is approximately 10 percentage points, which is approximately a 10% change.
The identification assumption underlying Eq. (1) is that by controlling for geographic area-year, we have controlled for any unmeasured characteristics of hospitals that would make reclassification endogenous—related to hospital staffing and patient outcomes. We acknowledge that this assumption is unlikely to hold particularly since our definition of geographic unit is larger than that consistent with the reclassification process. Ideally, we would like to compare hospitals on the urban fringe of say Chicago to similar hospitals, but there may be few hospitals in the same radius as the hospital of interest. More importantly, why would one hospital in a given radius of Chicago reclassify and another not? It is likely that these hospitals differ in unmeasured ways that affect staffing and patient outcomes.
An alternative is to control for hospital-specific effects. This is illustrated by the following equation:
equation(2)
RNijt=αi+β2(TREATi xPOSTt)+XitΓ+δt+uijtRNijt=αi+β2(TREATi xPOSTt)+XitΓ+δt+uijt
Note that in Eq. (2) there is a hospital-specific effect (αi) included in the model. 12 And now we control for just year (δt) and not geographic area-year (POSTt drops out of this model). Here, the identification assumption is that once we control for the hospital-specific effect, treatment status is exogenous. This is a more plausible assumption than that underlying Eq. (1). The threat in this case is that there are unmeasured, time-varying hospital characteristics that affect outcomes that do not operate through changes in reimbursement. The conditional nature of this statement is important to recognize. Reclassification will change reimbursement and this may affect hospital staffing and patient outcomes. This is not a problem and is in fact the identification we seek to exploit.
One way to assess the validity of the identification assumption underlying Eq. (2) is to investigate whether treatment and control hospitals have similar year effects prior to reclassification, but with only two observations per treatment hospital, one pre-reclassification and one post-reclassification, this approach is not feasible. However, we can expand the sample to include additional pre-reclassification years and doing so would allow us to investigate the hypothesis that outcomes of treatment and control hospitals have similar time trends in the absence of reclassification. Specifically, we use a sample that includes all pre-reclassification observations available in the data (1994–2000), which we describe in detail below, and estimate the following:
equation(3)
<img height="48" border="0" style="vertical-align:bottom" width="360" alt="View the MathML source" title="View the MathML source" src="http://origin-ars.els-cdn.com/content/image/1-s2.0-S0167629607001087-si3.gif">RNijt=αi+∑t=19941999βt(TREATi xδt)+XitΓ+δt+uijt
If the underlying identification assumption of our approach is valid, the coefficients on the interaction terms between the treatment group indicator and year effects should be zero. This would indicate that in the absence of reclassification, hospital resources (e.g., nurses) and patient outcomes would have similar time trends in treatment and control hospitals. In fact, estimates of Eq. (3) fail to reject a the 0.05 level the hypothesis that year effects are equal for treatment and control hospitals in 17 of 20 specifications. We describe these results in more detail below, but conclude that this specification test provides relatively strong support for our empirical approach.
To further bolster the research design, it is possible to include hospital fixed effects (αi) and geographic area-year effects (δjt), although in this case, year refers to the year of reclassification and not each year that data is observed. This specification is given by the following equation:
equation(4)
RNijt=αi+β2(TREATi xPOSTt)+XitΓ+δjt+uijtRNijt=αi+β2(TREATi xPOSTt)+XitΓ+δjt+uijt
In this specification, we control for unmeasured, hospital-specific effects (αi), and time-varying effects at the geographic area-year level. To be clear about this specification it helps to consider a case where there are only four hospitals, one treatment and three controls, and all are from the same area. Thus there are eight observations: 2 years (pre- and post-reclassification) of data for four hospitals. Our model would include the following six variables: four dummy variables identifying each hospital, the operating wage index, and a dummy variable indicating the post-reclassification year. In this case, the dummy variable indicating the year after reclassification is measuring time varying area-specific effects—time-varying effects common to all hospitals in this area (this is trivially true since there is only one area). The same basic logic can be expanded to many areas.
What makes Eq. (4) difficult to estimate is that the degrees of freedom are quickly exhausted. This specification is only feasible if there are a large number of hospitals in each area, which is generally not the case. However, if the unit of analysis is the patient in hospital (i), there are ample degrees of freedom since there are many observations per hospital. Therefore, Eq. (4) will only be estimated for models where we are examining the effect of Medicare reimbursement on patient outcomes.
3.3. Data
Three data sources were used in the analysis: the Prospective Payment System Payment Impact (PPS Impact) files from 1994 to 2001 from the Center for Medicare and Medicaid (CMS), the American Hospital Association Annual Surveys from 1994 to 2001, and the Nationwide Inpatient Sample (NIS) for 1994–2001 from the Healthcare Cost and Utilization Project (HCUP) of the Agency for Healthcare Research and Quality (AHRQ).
Our first task was to identify hospitals that were reclassified (or declassified). To accomplish this task, we used the PPS impact files. These files are maintained by CMS and used to calculate payments under Medicare's PPS. Most importantly, they provide information on the operating wage index, a key determinant of Medicare reimbursement rates, and whether a hospital has been geographically reclassified. The only other information we used from these files was the Medicare case mix index, which measures the average severity of illness. We merged the files from 1994 to 2001 so that we had a complete history of changes in the operating wage index of the hospital between 1994 and 2001. For all hospitals with non-missing observations (99% of sample), we use the administrative codes provided in the data to select those that were reclassified, or declassified, with respect to the operating wage index between 1994 and 2001. We refer to these as “switchers” and each “switcher” contributed 2 years of data: the year that reclassification (declassification) became effective, and year t − 2 (2 years prior). Thus, a hospital that was reclassified in year t, declassified in year t + 1 and reclassified in year t + 2 was not included. The reason for this restriction is that hospitals know the outcome of their reclassification petition 6 months prior to it taking effect. Thus, we thought it prudent to omit (in some analyses) the year prior to reclassification (declassification) because hospitals may have responded to the impending reclassification in that year even though their reimbursement remained unchanged. Note that a declassified hospital had to have been reclassified for at least 2 years to be included in the sample of “switchers.” For example, a hospital that was reclassified in year t and declassified in year t + 2 was included, but a hospital that was reclassified in year t and declassified in year t + 1 was not included (this hospital may be in the sample as a reclassified hospital if it had not changed status in years t − 2 and t − 1). Finally, a hospital could be a “switcher” in more than one period, although relatively few (15%) are in the sample more than once, and each instance of switching is treated as a separate unit in terms of specifying hospital fixed effects.
We identified approximately 927 “switchers” in the PPS files of which 598 were reclassified and 329 were declassified. For each switching hospital, we selected all other hospitals in the same geographic area of origin, or alternatively in the same geographic area of destination, in the year of the switch (reclassified or declassified). We refer to these as control hospitals. By geographic area of origin, we refer to the Medicare defined geographic area that hospital is actually located within, and the area of destination is the Medicare defined geographic area that was used to set reimbursement after reclassification. Each control hospital contributed 2 years of data, and to be a control, the hospital had to not have a change in its operating wage index during the period (2 years prior to switch and year switch became effective). Control hospitals may be in the sample multiple times. If there were no control hospitals in an area, we dropped the “switcher,” and this left 805 “switchers” with controls chosen on the basis of the originating geographic area, and 905 “switchers” with controls chosen on the basis of the destination area.
The second data source we used was the American Hospital Association Annual Surveys (AHA) from 1994 to 2001. Each year the American Hospital Association has conducted its Annual Survey of hospitals to collect information about the hospital's organizational structure, personnel, facilities and services, and financial performance. The response rate is over 80%. We used these data to obtain information about the following for each hospital:
•
number of beds in hospital unit and nursing home unit (if present);
•
ownership status (non-profit, for-profit);
•
teaching status;
•
nursing (RN, LPN) resources used in fiscal year, as measured by total facility (hospital and nursing home), full-time equivalents (FTEs);
•
physician resources used in fiscal year, as measured by total facility, full-time equivalents (FTEs);
•
total inpatient admissions in hospital unit and nursing home unit (if present) in fiscal year;
•
inpatient days in hospital unit and nursing home unit (if present) in fiscal year;
•
Medicare inpatient days in hospital and nursing home unit (if present) in fiscal year;
•
and total number of outpatient visits in fiscal year.
We restricted the sample to private (non-profit or for-profit), short-term general medical and surgical hospitals, and among this group, to hospitals that actually report (no imputation) information for the fiscal year of the survey. Since hospitals use different fiscal year definitions, we defined calendar year as the year with the greatest overlap with the fiscal year. These sample restrictions resulted in approximately 2900 hospitals per year.
It is important to note that nursing and physician resources refer to the total facility and not just the hospital unit. We are interested in the effect of changes in Medicare reimbursement on nursing resources per patient in the hospital unit. Therefore, in analyses of the effect of reclassification on nursing use, we control in a flexible way (i.e., several dummy variables) for the number of beds, admissions and inpatient days in both the hospital and nursing home units. Note that these variables are included only in the analyses of hospital use of nursing resources because the information on nurses and physicians is not reported separately for the hospital and nursing home unit.
Next, we merged the data from the PPS Impact files to the data from the AHA files. This resulted in the following number of “switchers” and “controls.”
•
342 switchers and 2870 controls using the geographic area of origin to choose controls;
•
and 393 switchers and 2178 controls using geographic area of destination to choose controls.
These are the samples of hospitals that we use to examine changes in nursing resources in response to a change in Medicare reimbursement. It is important to note that the decrease in the number of “switchers” from that identified solely from the PPS Impact data is driven entirely by the restrictions used to select the sample of hospitals from the AHA data.13
The final data source used in the analysis is the Nationwide Inpatient Sample (NIS) for 1994–2001. We use these data to obtain information about patient outcomes. The NIS is a public use file containing information on every discharge from a 20% sample of non-federal, short-term, general and other specialty hospitals in a sample of states. The number of states reporting information has grown over time. In 1994, 15 states participated, and in 1999, 19 states participated.14 While the number of states that have participated in HCUP has grown since 1999, this is the last year that is relevant to our analysis since we require a hospital to be present for 3 years, and 2001 is the last year of information from our other data sources. The NIS has information about: the age of the patient, state of residence, hospital characteristics (from AHA survey), the principal and secondary diagnoses, principal procedures, length of stay, and inpatient mortality.
We merged the patient discharge records from the NIS to the hospital level data using the AHA identification number. Unfortunately, relatively few “switchers” merge, and as a result the number of “switchers” for which we have patient information is 26. The primary reasons for this dramatic decrease are the limited number of states that participated in the NIS, and the 20% sample frame used to select hospitals in those states. Specifically, 64% of the switchers are lost because of the limited number of states that participated in NIS, and 27% are lost due to the hospital not being present in three consecutive years.15 There was a corresponding decrease in the number of control hospitals.
To address the small sample size problem, we selected another sample using a slightly different procedure. Switchers were defined the same as before – i.e., as a hospital that was reclassified or declassified – but instead of restricting the sample to observations in years t − 2 to t (year switch became effective), we used observations from any pre- and post-switching year between 1994 and 2001. This significantly increased the sample size because of the increase in the number of matches from the Nationwide Inpatient Sample. To see why, consider the following. Previously, if a hospital was reclassified in 1999, we selected observations of that hospital in 1997 and 1999, and there was a high probability that the hospital would not be in the NIS in either or both years. By expanding the years, we greatly increased our chances of finding a match. We defined control hospitals as any hospital that was never re(de)-classified, and used any observation of these hospitals between 1994 and 2001. This selection process also mandates that we construct separate analysis files for hospitals that are reclassified and those that are declassified because there may be observations in which a hospital is a post-reclassification year and simultaneously a pre-declassification year.
The disadvantage of this selection process is that we lose the narrowly defined pre- and post-reclassification research design. Under the alternative selection process, we compare pre- and post-reclassification changes that may be several years apart, and thus the assumptions underlying the fixed effects procedure may be less likely to hold. However, in most cases, we continue to have a relatively small study period of 3–5 years. The advantage is that our sample size increases markedly. After merging with the NIS, we are left with approximately 160 switching hospitals and between 330 and 460 control hospitals depending whether we are selecting control hospitals from the area of origin or area of destination.
For each hospital with patient level information, we limited the sample to patients between the ages of 40 and 80. We also restricted the patient records to those admitted for a limited number of illnesses (primary diagnoses), defined by ICD-9 codes. The purpose of using patient records was to examine outcomes that may be related to changes in hospital resource use due to changes in Medicare reimbursement. Since nursing resources are the largest single source of labor, we focused on outcomes that are particularly sensitive to nursing care (Needleman et al., 2005). These outcomes include urinary tract infections, pressure ulcers, secondary pneumonia, deep vein thrombosis, and failure to rescue. Identification of these outcomes using diagnostic and procedure codes is provided by Needleman et al. (2001). We also use inpatient mortality as an outcome. In addition, we limited the patient sample to those admitted for primary diagnoses for which the incidence of the outcomes of interest was not rare (<1%). To select these diagnoses, we examined all discharges in the HCUP data for 1998, and calculated the incidence of each outcome by primary diagnosis and ranked the outcomes. Based on this analysis, we selected all admissions for the following primary diagnoses (ICD-9 codes): Septicemia; volume depletion; ischemic heart disease; diseases of pulmonary circulation; other forms of heart disease; cerebrovascular disease; diseases of arteries, arterioles, and capillaries; diseases of veins, and lymphatics, and other diseases of the circulatory system; pneumonia and influenza; pneumoconioses and other lung diseases due to external agents; and other diseases of the respiratory system.16
4. Results
4.1. Descriptive statistics on treatment and control hospitals
Table 1 presents descriptive statistics of hospital-level characteristics, measured 2 years before reclassification, for the sample of hospitals used in the analysis. Separate statistics are calculated for “switchers’ and “controls.” Two sets of figures are shown: one for “switchers” and “controls” when “controls” are chose from the geographic area of origin; and the other when “controls” are chosen from the geographic area of destination. The number of “switchers” differs in the two samples because we drop “switchers” from areas that that do not have any “controls.”
Table 1.
Differences in characteristics between hospitals reclassified and hospitals not reclassified 2 years prior to reclassification
|
Variable |
Controls chosen from originating area
|
Controls chosen from destination area
|
||||||
|
|
Switchers |
Controls |
Dif. |
Adj. Dif. |
Switchers |
Controls |
Dif. |
Adj. Dif. |
|
Hospital unit |
||||||||
|
Beds |
132.2 |
87.92 |
44.28 |
36.46* |
150.3 |
226.0 |
−75.70 |
−80.13* |
|
Admissions/1000 |
5.495 |
3.373 |
2.122 |
1.670* |
6.289 |
9.662 |
−3.373 |
−3.501* |
|
Inpatient days/1000 |
26.24 |
16.58 |
9.660 |
7.412* |
30.81 |
52.30 |
−21.49 |
−21.26* |
|
Medicare share of inpatient days |
0.573 |
0.594 |
−0.021 |
−0.010 |
0.561 |
0.529 |
0.032 |
0.035* |
|
Medicare case mix |
1.246 |
1.156 |
0.090 |
0.079* |
1.270 |
1.364 |
−0.094 |
−0.117* |
|
|
||||||||
|
Total facility |
||||||||
|
Beds |
148.8 |
107.4 |
41.40 |
36.08* |
165.5 |
243.8 |
−78.30 |
−83.17* |
|
Admissions/1000 |
5.617 |
3.444 |
2.173 |
1.720* |
6.404 |
9.840 |
−3.436 |
−3.573* |
|
Inpatient days/1000 |
31.72 |
22.98 |
8.740 |
7.381* |
35.76 |
57.69 |
−21.93 |
−21.84* |
|
Medicare share of inpatient days |
0.545 |
0.498 |
0.047 |
0.035* |
0.540 |
0.517 |
0.023 |
0.025* |
|
RN |
149.2 |
95.02 |
54.18 |
45.59* |
171.98 |
290.1 |
−118.1 |
−120.8* |
|
RN/(Tot Fac Admissions/1000) |
27.41 |
29.29 |
−1.880 |
−1.250* |
27.51 |
29.03 |
−1.520 |
−1.420* |
|
LPN |
30.88 |
21.24 |
9.640 |
8.247* |
31.15 |
32.22 |
−1.070 |
−6.399* |
|
LPN/(Tot Fac Admissions/1000) |
7.089 |
9.324 |
−2.235 |
−2.176* |
6.469 |
4.714 |
1.755 |
1.442* |
|
PHY |
7.564 |
7.122 |
0.442 |
1.771 |
8.883 |
25.00 |
−16.12 |
−11.95* |
|
PHY/(Tot fac admissions/1000) |
1.582 |
1.616 |
−0.034 |
0.102 |
1.471 |
1.716 |
−0.245 |
−0.125 |
|
For profit |
0.184 |
0.099 |
0.085 |
0.068* |
0.178 |
0.130 |
0.048 |
0.022 |
|
Teaching |
0.029 |
0.020 |
0.009 |
0.011 |
0.041 |
0.124 |
−0.083 |
−0.051* |
|
High technology |
0.216 |
0.115 |
0.101 |
0.099* |
0.265 |
0.444 |
−0.180 |
−0.209* |
|
Observations |
342 |
2870 |
|
|
393 |
2178 |
|
|
Notes: Switchers are hospitals that changed geographic location with respect to assignment of operating wage index used to compute Medicare reimbursement. Controls are those that did not change geographic location with respect to operating wage index used to compute Medicare reimbursement. Adjusted differences control for area-year effects. In other words, adjusted differences between switchers and controls pertain to hospitals in the same area, defined prior to reclassification, and year. Hospital unit and total facility Medicare inpatient share is missing in a few cases. Hospital unit – origin area N: 321 switchers, 2712 controls; destination area N: 372 switchers, 2067 controls. Total facility – origin area observations: 337 switchers, 2816 controls; destination area N: 386 switchers, 2152 controls.
*
Statistically significant difference (p < 0.05); standard error of difference was calculated under the assumption that observations were not independent within year and geographic area.
As can be seen in Table 1, “switchers” are larger than hospitals in their area of origin and smaller than hospitals in their area of destination. The average “switcher” has approximately 36% (41%) more beds than other hospitals their area of origin and 80% (35%) fewer beds than hospitals in their area of destination. “Switchers” also have a higher Medicare Case Mix index than hospitals in their area of origin and a lower Medicare Case Mix index than hospitals in their area of destination. These differences between switchers and controls are consistent with the process of geographical classification, which is mainly a way for rural and urban fringe hospitals with unusually high labor costs to obtain greater reimbursement. So it appears that it is the larger hospitals with sicker patients in rural and urban fringe areas that have the high labor costs. Interestingly, “switchers” use fewer LPN resources per admission than hospitals in the area or origin, but more LPN resources per admission than hospitals in the area of destination. “Switchers” also use fewer RN resources per admission than other hospitals in the area of origin and destination.
Overall, the figures in Table 1 demonstrate that “switchers” are different from other hospitals in both their area of origin and their area of destination. So it is unlikely that simply controlling for area of origin, or area of destination, is sufficient to establish (conditional) exogeneity of “switcher” status. A more plausible approach would be to control for hospital-specific effects. Below, we present results from both approaches.
4.2. Effects of geographic reclassification on hospitals’ use of resources
Table 2 presents regression estimates of the effects of Medicare reimbursement on hospital use of resources, as measured by admissions, inpatient days, and RN and LPN nursing resources. RN and LPN quantities are measured as full time equivalents (FTE) at the facility (hospital and nursing home units) level. For these outcomes, we need to control for “quantity” because our interest is in nurses per patient (or per admission). Accordingly, we control for differences in the number of beds, admissions and patients in both the hospital and nursing home units.
Table 2.
Estimates of the Effect of geographical reclassification on hospital characteristics
|
Variable |
Controls chosen from originating area
|
Controls chosen from destination area
|
||||||
|
Hospital unit |
Reclassifier |
Declassifier |
Reclassifier |
Declassifier |
Reclassifier |
Declassifier |
Reclassifier |
Declassifier |
|
Admissions/1000 |
−0.072 (0.089) |
0.120 (0.175) |
0.020 (0.111) |
0.176 (0.160) |
−0.020 (0.176) |
0.269 (0.228) |
−0.194 (0.165) |
−0.034 (0.172) |
|
Inpatient Days/1000 |
−0.770+ (0.057) |
0.396 (1.006) |
−0.218 (0.539) |
0.533 (1.038) |
−0.056 (0.886) |
1.987 (1.177) |
−0.807 (0.664) |
0.875 (0.972) |
|
Medicare share of inpatient days |
0.001 (0.008) |
0.007 (0.012) |
−0.001 (0.011) |
0.012 (0.016) |
0.009 (0.008) |
−0.011 (0.010) |
−0.003 (0.010) |
−0.0002 (0.013) |
|
|
||||||||
|
Total facility |
||||||||
|
RN |
−3.801 (3.656) |
−5.154 (5.247) |
−4.771 (3.918) |
3.639 (5.096) |
−0.568 (5.183) |
9.606 (7.143) |
−6.550 (5.388) |
5.637 (6.212) |
|
LPN |
−1.320 (0.847) |
−1.530 (1.150) |
−0.771 (1.303) |
−1.227 (1.423) |
0.418 (1.461) |
0.237 (1.392) |
0.358 (1.457) |
−0.080 (1.566) |
|
Area-year fixed effects |
Yes |
No |
Yes |
No |
||||
|
Hospital fixed effects |
No |
Yes |
No |
Yes |
||||
|
Year fixed effects |
No |
Yes |
No |
Yes |
||||
|
Observations |
6424 |
6424 |
5142 |
5142 |
||||
|
Number of hospitals |
3212 |
3212 |
2571 |
2571 |
Notes: All regressions control for Medicare case mix and for teaching, hi-tech and for-profit status. In addition, the RN and LPN regressions also control for hospital unit and nursing home beds, admissions and inpatient days; and outpatient and missing outpatient visits. Medicare share has missing observations. Medicare share observations and number of hospitals, respectively—Origin area: 6095, 3186; Destination area: 4904, 2546. *Statistically significant difference (p < 0.05); +Statistically significant difference (0.05 ≤ p < 0.10); standard errors were calculated under the assumption that observations were not independent within hospital.
Our preferred estimates are those that hold constant hospital-specific effects, so we limit the discussion to these estimates. None of the hospital, fixed effects estimates in Table 2 are statistically significant, and most are also small in magnitude. For example, hospitals that were reclassified, and which therefore experienced approximately a 10% increase in Medicare reimbursement, employed approximately five (4.771) to seven (6.550) fewer nurses, which is a three to 4% decline. Hospitals that were declassified, and who experienced decreases in Medicare reimbursement of approximately 10%, increased the number of nurses by four (3.639) to six (5.637), or 2% to 4%. Other estimates in the table have magnitudes with similarly small relative effect sizes.17 These estimates are inconsistent with theoretical predictions of a positive relationship between reimbursement levels and nurse resources (quality), and reimbursement and quantity of patients.
We re-estimated the models of Table 2 using the alternative sample selection procedure that yielded a larger sample of hospitals: between 549 and 611 switchers and 1300 and 2070 controls depending on whether we selected controls from the area of origin or the area of destination. Note that the number of control hospitals decreased relative to the initial sample selection rule because we did not allow hospitals to enter the analysis multiple times. Estimates using the larger sample are shown in Appendix Table A2. There are no substantial differences between the estimates in Table 2 and Appendix Table A2. None of the estimates are statistically significant and all are relatively small in magnitude.
As we noted above, we also assessed the validity of the identification assumption underlying the estimates in Table 2 (and Appendix Table A2). Specifically, we tested whether pre-reclassification trends in the five-dependent variables were equal between treatment and control hospitals. For this analysis, we used the larger sample of hospitals because we required more than 1 year of pre-reclassification data. Twenty regressions were estimated: five-dependent variables (admissions, inpatient days, Medicare share of inpatient days, RN, and LPN), two treatments (reclassified and declassified), and two control groups (origin and destination). In only three of the 20 cases could we reject the null hypothesis of equal pre-treatment trends. All three rejections involved the use of control hospitals from the destination area and two of the rejections pertained to admissions and one pertained to inpatient days. We believe these results provide relatively strong support that our research design is valid, particularly in cases where we use controls from the area of origin.
One hypothesis we tested was whether the effect of changes in Medicare reimbursement differed by the share of Medicare patients. This is similar to several other previous papers who use an explicit Medicare “bite” variable (Gruber, 1994, Cutler, 1998, Staiger and Gaumer, 1995 and Shen, 2003). Hospitals with a relatively large share of Medicare patients would receive larger absolute increases in revenues and may respond differently than hospitals with smaller shares of Medicare patients. In addition, some theoretical model such as Glazer and McGuire (2002) suggest that hospital responses to changes in reimbursement may differ by share of Medicare patients because of joint production. To examine this issue, we allowed the effects of reclassification to differ by the initial share of Medicare patients. We calculated the share of Medicare patients in year t − 2 and used that share to classify hospitals into three categories: 0–49%, 50–69%, and 70% or more. Table 3a and Table 3b present the estimates: Table 3a shows estimates when “controls” were chosen from the geographic area of origin, and Table 3b shows estimates when “controls” were chosen from the geographic area of destination. Again, we focus the discussion on the models that control for hospital-specific fixed effects (other models not shown).
Table 3a.
Estimates of the effect of geographical reclassification
|
Variable |
Medicare share 0–49% |
Medicare share 50–69% |
Medicare share > =70% |
Medicare share 0–49% |
Medicare share 50–69% |
Medicare share > =70% |
|
Hospital unit |
Reclassifier |
Reclassifier |
Reclassifier |
Declassifier |
Declassifier |
Declassifier |
|
Admissions/1000 |
−0.337 (0.445) |
0.044 (0.063) |
−0.061 (0.172) |
0.412 (0.569) |
0.149 (0.092) |
−0.152 (0.140) |
|
Inpatient days/1000 |
−1.674 (1.962) |
−0.119 (0.494) |
−0.240 (0.631) |
−0.895 (1.685) |
0.306 (0.731) |
1.108 (2.911) |
|
|
||||||
|
Total facility |
||||||
|
RN |
−11.35 (13.90) |
−4.104 (3.636) |
−1.329 (5.821) |
6.247 (14.92) |
−0.389 (5.476) |
10.17 (7.720) |
|
LPN |
0.373 (3.398) |
−0.660 (1.592) |
−1.746 (1.755) |
−2.485 (2.427) |
−1.463 (1.749) |
−0.661 (4.287) |
|
Area-year fixed effects |
No |
No |
||||
|
Hospital fixed effects |
Yes |
Yes |
||||
|
Year fixed effects |
Yes |
Yes |
||||
|
Observations |
6424 |
6424 |
||||
|
Number of hospitals |
3212 |
3212 |
Notes: All regressions control for Medicare case mix, and for teaching, hi-tech and for-profit status. In addition, the RN and LPN regressions also control for hospital unit and nursing home beds, admissions and inpatient days; and outpatient and missing outpatient visits. Medicare share refers to the share of hospital unit inpatient days attributable to Medicare. *Statistically significant difference (p < 0.05); +statistically significant difference (0.05 ≤ p < 0.10); standard errors were calculated under the assumption that observations were not independent within hospital. All estimates pertain to periods (t − 2) and (t), where t is the year of reclassification. Medicare share has only 6095 observations for 3186 hospitals. By Medicare Share of Inpatient Days on Hospital Characteristics—Controls Chosen from Originating Area.
Table 3b.
Estimates of the effect of geographical reclassification on hospital characteristics by Medicare share of inpatient days on hospital characteristics—controls chosen from destination area
|
Variable |
Medicare share 0–49% |
Medicare share 50–69% |
Medicare share > =70% |
Medicare share 0–49% |
Medicare share 50–69% |
Medicare share > =70% |
|
Hospital unit |
Reclassifier |
Reclassifier |
Reclassifier |
Declassifier |
Declassifier |
Declassifier |
|
Admissions/1000 |
−0.603 (0.420) |
−0.100 (0.189) |
−0.383* (0.191) |
0.143 (0.370) |
−0.003 (0.210) |
−0.429* (0.180) |
|
Inpatient days/1000 |
−2.050 (1.891) |
−0.648 (0.655) |
−0.817 (0.887) |
0.480 (1.608) |
1.167 (0.940) |
−0.226 (2.891) |
|
|
||||||
|
Total facility |
||||||
|
RN |
−9.064 (15.39) |
−5.162 (5.844) |
−5.525 (7.100) |
16.43 (13.78) |
0.649 (7.134) |
6.041 (8.160) |
|
LPN |
1.334 (3.190) |
0.515 (1.831) |
−0.751 (2.090) |
0.558 (2.211) |
−0.139 (1.909) |
−1.096 (4.248) |
|
Area-year fixed effects |
No |
No |
||||
|
Hospital fixed effects |
Yes |
Yes |
||||
|
Year fixed effects |
Yes |
Yes |
||||
|
Observations |
5142 |
5142 |
||||
|
Number of hospitals |
2571 |
2571 |
Notes: All regressions control for Medicare casemix, and for teaching, hi-tech and for-profit status. In addition, the RN and LPN also control for hospital unit and nursing unit beds, admissions and inpatient days; and outpatient and missing outpatient visits. Medicare share refers to the share of hospital unit inpatient days attributable to Medicare. All estimates pertain to periods (t − 2) and (t), where t is the year of reclassification. Medicare share has only 4904 observations on 2546 hospitals, 188 of which have missing share in one of the periods.
*
Statistically significant difference (p < 0.05); + indicates a statistically significant difference (0.05 ≤ p < 0.10); standard errors were calculated under the assumption that observations were not independent within hospital.
Estimates in Table 3a and Table 3b do not indicate that hospital responses to changes in Medicare reimbursement differed significantly by share of Medicare patients. First, the overwhelming majority of estimates are not statistically different from zero, and second, most estimates are also relatively small—less than two or 3% in relative terms. The precision of the estimates decreases, however, making it difficult to detect small effects reliably. For example, for hospitals with a small share of Medicare patients (0–49%), we observe changes in RNs in the 10% range, but these estimates are statistically insignificant. Overall, there is little evidence that hospital responses to reclassification differed by the share of Medicare patients they treat. In fact, the signs of the coefficients are usually the opposite of that predicted by theory; reclassification and increases in reimbursement are associated with decreases in admissions and use of nursing resources, and declassification and decreases in reimbursement are associated with increases in nursing resources.
In sum, the results to this point show that significant changes in Medicare reimbursement had little effect on hospital resource use as measured by admissions, inpatient days, and nurse resources. To assess whether the length of time that a hospital was reclassified (declassified) mattered, we allowed the effect of reclassification (declassification) to differ by whether the hospital had a changed status for 1 to 2 years or three or more years. These results (not shown) are consistent with those presented earlier: changes in reimbursement are not significantly related to changes in resource use. There was no dose–response effect associated with being reclassified for a longer period.
4.3. Effects of geographic reclassification on patient outcomes
We now turn to the effect of Medicare reimbursement on patient outcomes. As noted, we focus on the effects of Medicare reimbursement on an expanded set of outcomes relative to the previous literature and in particular on nurse-sensitive patient outcomes. These are outcomes that are considered to be particular sensitive to changes in nursing resources and include: secondary pneumonia, deep vein thrombosis, failure-to-rescue, pressure ulcers, and urinary tract infections. We chose these outcomes because nurses represent approximately 30% of hospital operating budgets and are the largest department in hospitals. Therefore, nurse resources and outcomes sensitive to changes in this resource are likely to be affected by changes in Medicare reimbursement. In addition to these patient outcomes, we examine in-hospital mortality and length of stay. Previously, we found little evidence that changes in Medicare reimbursement are associated with changes in nursing resources, or changes in hospital volume. This suggests that we should find no association between changes in Medicare reimbursement and nurse-sensitive outcomes, or even mortality. However, we have not examined changes in other resources in response to changes in Medicare reimbursement, and these potential changes may affect outcomes.
Table 4 presents the means of patient outcomes in year (t − 2) for “switcher” and “control” hospitals. We divide patients into two groups: those between the ages of 40 and 64, which are predominantly private payer, and those 65–80 who are all covered by Medicare. The division of the sample into private (age 40–65) and public (age 65–80) patients allows us to investigate whether changes in Medicare reimbursement affected private as well as public (i.e. Medicare) patients. With one exception, there are no statistically significant differences in patient outcomes between the two types of hospitals. The lone exception is the rate of failure to rescue among patients aged 65–80, which is higher among “control” hospitals in the geographic area of origin.
Table 4.
Differences in patient outcomes between hospitals reclassified and hospitals not reclassified 2 years prior to reclassification
|
Variable |
Controls chosen from originating area
|
Controls chosen from destination area
|
||||||
|
|
Switchers |
Controls |
Dif. |
Adj. Dif. |
Switchers |
Controls |
Dif. |
Adj. Dif. |
|
Ages 40–64 |
||||||||
|
Mortality |
0.030 |
0.031 |
−0.001 |
−0.001 |
0.034 |
0.037 |
−0.003 |
−0.006 |
|
Pneumonia |
0.025 |
0.021 |
0.004 |
0.005 |
0.033 |
0.036 |
−0.003 |
−0.003 |
|
DVT/PE |
0.007 |
0.009 |
−0.002 |
−0.001 |
0.008 |
0.011 |
−0.003 |
−0.001 |
|
Failure to rescue |
0.107 |
0.103 |
0.004 |
−0.001 |
0.114 |
0.124 |
−0.010 |
−0.021 |
|
Urinary tract infection |
0.042 |
0.044 |
−0.002 |
−0.003 |
0.040 |
0.045 |
−0.005 |
−0.004 |
|
Pressure ulcers |
0.017 |
0.015 |
0.002 |
−0.0004 |
0.015 |
0.021 |
−0.006 |
−0.007 |
|
Length of stay (natural log) |
1.264 |
1.261 |
0.003 |
−0.000 |
1.380 |
1.445 |
−0.065 |
−0.068 |
|
Observations |
8828 |
15960 |
|
|
13963 |
48766 |
|
|
|
Number of hospitals |
26 |
94 |
|
|
26 |
65 |
|
|
|
|
||||||||
|
Ages 65–80 |
||||||||
|
Mortality |
0.052 |
0.058 |
−0.006 |
−0.006 |
0.060 |
0.066 |
−0.006 |
−0.005 |
|
Pneumonia |
0.035 |
0.033 |
0.002 |
0.005 |
0.038 |
0.048 |
−0.010 |
−0.009 |
|
DVT/PE |
0.006 |
0.008 |
−0.002 |
−0.001 |
0.008 |
0.011 |
−0.003 |
−0.001 |
|
Failure to rescue |
0.119 |
0.171 |
−0.052 |
−0.051* |
0.150 |
0.173 |
−0.023 |
−0.014 |
|
Urinary tract infection |
0.080 |
0.076 |
0.004 |
0.003 |
0.073 |
0.087 |
−0.014 |
−0.012 |
|
Pressure ulcers |
0.019 |
0.020 |
−0.001 |
−0.002 |
0.022 |
0.030 |
−0.008 |
−0.009 |
|
Length of stay (natural log) |
1.420 |
1.416 |
0.004 |
0.026 |
1.533 |
1.612 |
−0.079 |
−0.049 |
|
Observations |
17135 |
32799 |
|
|
24664 |
78388 |
|
|
|
Number of hospitals |
26 |
94 |
|
|
26 |
65 |
|
|
Notes: Switchers are hospitals that changed geographic location with respect to assignment of operating wage index used to compute Medicare reimbursement. Controls are those that did not change geographic location with respect to operating wage index used to compute Medicare reimbursement. Adjusted differences control for area-year effects. In other words, differences between switchers and controls pertain to hospitals in the same area, defined prior to reclassification, and year. The failure-to-rescue outcome has 6% of the observations of the other outcomes because its risk pool is smaller than it is for the rest of the outcomes.
*
Statistically significant difference (p < 0.05); standard error of difference was calculated under the assumption that observations were not independent within year and geographic area.
Estimates of the effect of geographic reclassification on patient outcomes are presented in Table 5. Here, our preferred model is that which controls for hospital-specific effects and area-year effects. We begin the discussion with the results obtained when controls were chosen from area of origin. Patient outcomes in hospitals that were reclassified and which received an increase in reimbursement showed both improvement and deterioration. There was a significant increase (26% and 36%, respectively) in the incidences of pressure ulcers and failure-to-rescue among those 65–80. However, there was a decrease (33%) in the incidence of DVT/PE among those 65–80. For patients in hospitals that were declassified, and who received lower Medicare reimbursement, there was an increase (50% and 100%, respectively) in mortality and failure-to-rescue among those 40–64.
Table 5.
Estimates of the effect of geographical reclassification on patient outcomes
|
Patient outcome |
Controls chosen from originating area
|
Controls chosen from destination area
|
||||
|
Ages 40–64 |
Reclassifier |
Declassifier |
(N) |
Reclassifier |
Declassifier |
(N) |
|
Ages 65–80 |
||||||
|
Mortality |
−0.001 (0.004) |
0.015* (0.005) |
50,713 |
0.001 (0.004) |
−0.003 (0.004) |
131,095 |
|
Pneumonia |
−0.004 (0.004) |
−0.0004 (0.006) |
50,730 |
0.002 (0.003) |
−0.004 (0.009) |
131,098 |
|
DVT/PE |
−0.000 (0.001) |
−0.003 (0.003) |
50,730 |
0.001 (0.001) |
0.002 (0.003) |
131,098 |
|
Failure to rescue |
−0.009 (0.044) |
0.090* (0.028) |
2,668 |
−0.022 (0.028) |
0.032 (0.046) |
8,654 |
|
Urinary tract infection |
0.005 (0.006) |
−0.009 (0.007) |
50,730 |
−0.0004 (0.004) |
−0.003 (0.005) |
131,098 |
|
Pressure ulcers |
0.003 (0.002) |
0.005 (0.003) |
50,730 |
0.003 (0.003) |
−0.001 (0.004) |
131,098 |
|
|
||||||
|
Ages 65–80 |
||||||
|
Mortality |
0.003 (0.004) |
0.013 (0.011) |
101,457 |
0.002 (0.003) |
−0.012* (0.003) |
213,271 |
|
Pneumonia |
0.0003 (0.006) |
0.008 (0.007) |
101,511 |
0.002 (0.003) |
−0.003 (0.006) |
213,269 |
|
DVT/PE |
−0.002* (0.001) |
0.002 (0.002) |
101,511 |
0.0002 (0.001) |
0.002+ (0.001) |
213,269 |
|
Failure to rescue |
0.044+ (0.023) |
−0.017 (0.057) |
7,193 |
−0.021 (0.024) |
−0.052+ (0.030) |
17,971 |
|
Urinary tract infection |
−0.003 (0.006) |
−0.004 (0.012) |
101,511 |
0.005 (0.006) |
0.008 (0.007) |
213,269 |
|
Pressure ulcers |
0.005* (0.002) |
−0.004 (0.003) |
101,511 |
0.002 (0.003) |
−0.0003 (0.006) |
213,269 |
|
Area-year fixed effects |
Yes |
Yes |
||||
|
Hospital fixed effects |
Yes |
Yes |
||||
|
Year fixed effects |
No |
No |
||||
|
Number of hospitals |
120 |
91 |
Notes: All regressions control for teaching, profit and hi-tech status; bed size, Medicare case mix index and ICD-9-CM codes at the three-digit level, of which there are 67 in the ages 40–64, originating area; 69 in the ages 40–64, destination area; 71 in the ages 65–80, originating area; and 69 in the ages 65–80, destination area. All estimates pertain to periods (t − 2) and (t), where t is the year of reclassification.
*
Statistically significant difference (p < 0.05); + indicates a statistically significant difference (0.05 ≤ p < 0.10); standard errors were calculated under the assumption that observations were not independent within hospital.
Estimates in the right panel of Table 5 are obtained using “controls” selected from the area of destination. In hospitals that were reclassified and who received an increase in reimbursement, there were no statistically significant changes in patient outcomes. In hospitals that were declassified, there was a decrease (approximately 30%) in the incidences of mortality and failure-to-rescue, and an increase (25%) in DVT/PE among those 65–80.
Overall estimates in Table 5 provide limited and mixed evidence as to the effect of Medicare reimbursement on patient outcomes. Increases in reimbursement are associated with both better and worse outcomes, and estimates are sensitive to the choice of “controls.” Given earlier results from the tests of the identification assumption, our preference is to emphasize results that use controls chosen from the area of origin. However, limiting discussion to these results, still finds that increases (decreases) in Medicare reimbursement are associated with both better and worse patient outcomes.
We re-estimated the models underlying Table 5 using the larger set of hospitals selected by the alternative sample selection process. In this case, we have approximately 160 hospitals that were re- or declassified, and between 330 and 460 control hospitals. Estimates obtained using these samples are listed in Appendix Table A3. Estimates obtained when controls were selected from the area of origin are for the most part not statistically significant and small. Among hospitals that were reclassified, increases in Medicare reimbursement were associated with a significant improvement (10–20%) in rates of pressure ulcers for both Medicare (65–80) and non-Medicare (40–64) patients. For hospitals that were declassified, decreases in Medicare reimbursement were associated with shorter lengths of stay (2.6%) and higher rates (20%) of pressure ulcers. While this evidence appears to be consistent with theoretical predictions, estimates obtained using controls from the area of destination do not confirm this finding. These estimates indicate that patient outcomes in reclassified hospitals worsened (urinary tract infections) and improved (pressure ulcers). Patient outcomes in declassified hospitals generally worsened (pneumonia, pressure ulcers), as predicted by theory, but length of stay increased approximately 2.5%. In sum, estimates in Appendix Table A3 suggest that changes in Medicare reimbursement had little systematic effect on patient outcomes.
5. Conclusions
In this paper, we examined the association between changes in Medicare reimbursement and hospital resource use and patient outcomes. Specifically, we studied the effect of changes in the Medicare operating wage index caused by geographic reclassification on hospital admissions, inpatient days and nursing resources, and on patient outcomes including length of stay. For patient outcomes, we focused on nurse-sensitive measures – outcomes that are considered to be sensitive to changes in the quantity of nursing care – and in-hospital mortality. Our objective was to obtain estimates of the associations of interest that could plausibly be given a causal interpretation. Accordingly, we used a research design that controlled for unmeasured hospital-specific effects that are potentially correlated with geographic reclassification and the outcomes of interest. And for the patient-level analyses, we included additional controls for unmeasured, time-varying geographic area effects.
Theoretical models in the literature (e.g., Chalkley and Malcomson, 2000, Hodgkin and McGuire, 1994 and Glazer and McGuire, 2002) predict that an increase in Medicare reimbursement would be associated with an increase in hospital admissions of Medicare patients (i.e., quantity), an increase in nursing resources (i.e., quality) and length of stay, and an improvement in patient outcomes. In contrast to these predictions, our reading of the evidence leads us to conclude that changes in Medicare reimbursement had no effect on hospital use of resources and patient outcomes.
The purpose of geographical reclassification is to provide a hospital with additional revenue to pay for labor costs that are higher than average. The conventional wisdom is that geographical reclassification is a necessary way to address problems in Medicare reimbursement that penalizes hospitals that draw employees from labor markets that have unusually high costs relative to other hospitals in their classification area. These unusually high labor costs put financial pressure on these hospitals that threaten the quality of care and patient outcomes. Geographic reclassification remedies this problem, so it is intended to affect the quality of care (resource use) and patient outcomes. Thus, we would expect increases (decreases) in Medicare revenue to increase (decrease) the quality of care and improve (worsen) patient outcomes. We did not find this. This suggests that geographic reclassification is not serving its intended purpose.
More generally, our results also suggest that the link between Medicare reimbursement and patient outcomes is not strong. We examined changes in revenue on the order of magnitude of 10%, which is not a trivial change, and found no evidence that patient outcomes were affected. These results are consistent with strategic government behavior. If reimbursement levels have a relatively weak effect on patient outcomes, it makes fiscal and public health sense for the government to ratchet those payment levels down to exploit the weak tradeoff between lower payments and patient care and outcomes. Eventually, inadequate payments must have some adverse effect, but perhaps the level of payments in our data is not in that range.
There are several explanations for these finding besides study limitations that we discuss below. First, patients may not be able to perceive quality, so hospital administrators may not have an incentive to improve quality as reimbursement increases; an improvement in quality will not affect patient demand and the hospital will not attract more patients by improving quality. Thus, the administrator can use the additional revenue to serve other objectives, for example, by raising employee and administrator wages. This may be particularly important in the case of Medicare since there is no third party payer that may have an incentive, and a much greater ability, to determine quality. It may be particularly difficult for individual patients to identify true quality. Second, the administrator, even those of non-profit hospitals, may not value quality, which would reinforce the previous mechanism. The number of for-profit hospitals in this study was too small to examine whether hospital responses differed by ownership status, but other evidence suggests little difference in hospital behavior by ownership status (Sloan et al., 2001). Third, the change in revenue due to reclassification may be viewed as temporary and therefore not acted upon, although hospitals that were reclassified for three or more years did not respond differently than hospitals reclassified for fewer years.
Study limitations may also explain the results. Our measure of nursing resources comes from the American Hospital Association Annual Surveys and is a crude measure that may contain significant measurement error. We also had a selected sample of hospitals – those that were geographically reclassified for reimbursement purposes – and as shown above these hospitals are larger and serve sicker patients than other hospitals in their area, but are smaller and have less sick patients than hospitals in the reclassified area. Moreover, because of data limitations, we had relatively few hospitals in the patient level analyses. So our findings, if valid, may still not be applicable to a large portion of hospitals.
Acknowledgements
Partial funding for this research was provided by the Robert Wood Johnson Foundation. The authors thank Darius Lakdawalla, Kosali Simon, two referees and an Associate Editor of this journal, participants of the Spring 2005 meeting of the NBER Health Economics Program, and seminar participants at Cornell-Weill College of Medicine for their helpful suggestions.
Appendix A.
See Table A1, Table A2 and Table A3.
Table A1.
Large Sample of Hospitals: Differences in Characteristics Between Hospitals Reclassified and Hospitals Not Reclassified In Years Prior to Reclassification
|
Variable |
Controls chosen from originating area
|
Controls chosen from destination area
|
||||||
|
|
Switchers |
Controls |
Dif. |
Adj. Dif. |
Switchers |
Controls |
Dif. |
Adj. Dif. |
|
Hospital unit |
||||||||
|
Beds |
131.5 |
105.4 |
26.10 |
48.26* |
142.9 |
207.5 |
−64.60 |
−77.73* |
|
Admissions/1000 |
5.304 |
4.279 |
1.025 |
2.224* |
5.942 |
9.012 |
−3.070 |
−3.198* |
|
Inpatient Days/1000 |
26.28 |
21.65 |
4.450 |
10.90* |
29.61 |
47.29 |
−17.68 |
−19.83* |
|
Medicare share of inpatient days |
0.562 |
0.577 |
−0.016 |
−0.012* |
0.558 |
0.533 |
0.025 |
0.034* |
|
Medicare Case Mix |
1.244 |
1.184 |
0.060 |
0.100* |
1.257 |
1.350 |
−0.093 |
−0.113* |
|
|
||||||||
|
Total facility |
||||||||
|
Beds |
147.9 |
125.3 |
22.60 |
46.40* |
158.2 |
225.8 |
−67.60 |
−81.55* |
|
Admissions/1000 |
5.403 |
4.360 |
1.043 |
2.262* |
6.036 |
9.172 |
−3.136 |
−3.262* |
|
Inpatient days/1000 |
31.54 |
28.11 |
3.430 |
10.24* |
34.50 |
52.91 |
−18.410 |
−20.85* |
|
Medicare share of inpatient days |
0.524 |
0.485 |
0.039 |
0.039* |
0.527 |
0.508 |
0.023 |
0.021* |
|
RN |
145.5 |
124.3 |
21.20 |
61.32* |
163.5 |
265.9 |
−102.4 |
−110.4* |
|
RN/(Tot Fac Admissions/1000) |
27.63 |
30.07 |
−2.440 |
−1.283* |
27.72 |
29.04 |
−1.320 |
−1.727* |
|
LPN |
28.88 |
20.60 |
8.280 |
9.482* |
29.36 |
30.39 |
−1.030 |
−6.111* |
|
LPN/(Tot Fac Admissions/1000) |
7.203 |
8.928 |
−1.725 |
−3.071* |
6.755 |
5.438 |
1.317 |
1.319* |
|
PHY |
7.918 |
11.23 |
−3.312 |
3.974* |
8.966 |
20.93 |
−11.96 |
−9.669* |
|
PHY/(Tot Fac Admissions/1000) |
1.422 |
1.941 |
−0.519 |
−0.108 |
1.396 |
1.764 |
−0.368 |
−0.246* |
|
For profit |
0.180 |
0.104 |
0.076 |
0.058* |
0.165 |
0.148 |
0.017 |
0.001 |
|
Teaching |
0.038 |
0.043 |
−0.005 |
0.024* |
0.043 |
0.109 |
−0.066 |
−0.047* |
|
High technology |
0.222 |
0.169 |
0.053 |
0.107* |
0.243 |
0.417 |
−0.174 |
−0.214* |
|
Observations |
2033 |
7187 |
|
|
2178 |
10618 |
|
|
|
Number of hospitals |
549 |
1300 |
|
|
611 |
2070 |
|
|
Notes: Switchers are hospitals that changed geographic location with respect to assignment of operating wage index used to compute Medicare reimbursement. Controls are those that did not change geographic location with respect to operating wage index used to compute Medicare reimbursement. Adjusted differences control for area-year effects. In other words, adjusted differences between switchers and controls pertain to hospitals in the same area, defined prior to reclassification, and year. Hospital unit and total facility Medicare inpatient share is missing in a few cases. Hospital unit – origin area N: 1921 switcher observations, 6767 control observations – destination area: 2069 switcher observations, 10065 control observations; total facility Medicare inpatient days – origin area: 1989 switcher observations, 7010 control observations – destination area: 2130 switcher observations, 10401control observations.
*
Statistically significant difference (p < 0.05); standard error of difference was calculated under the assumption that observations were not independent within year and geographic area.
Table A2.
Estimates of the Effect of Geographical Reclassification on Hospital Characteristics
|
Variable |
Controls chosen from originating area
|
Controls chosen from destination area
|
||||||
|
Hospital unit |
Large sample
|
Small sample
|
Large sample
|
Small sample
|
||||
|
|
Reclassifier |
Declassifier |
Reclassifier |
Declassifier |
Reclassifier |
Declassifier |
Reclassifier |
Declassifier |
|
Admissions/1000 |
0.038 (0.058) |
0.012 (0.055) |
0.020 (0.111) |
0.176 (0.160) |
−0.181 (0.117) |
−0.226* (0.084) |
−0.194 (0.165) |
−0.034 (0.172) |
|
Inpatient days/1000 |
−0.275 (0.337) |
−0.015 (0.373) |
−0.218 (0.539) |
0.533 (1.038) |
−0.333 (0.404) |
−0.010 (0.467) |
−0.807 (0.664) |
0.875 (0.972) |
|
Medicare share of inpatient days |
0.001 (0.006) |
0.005 (0.007) |
−0.001 (0.011) |
0.012 (0.016) |
0.006 (0.005) |
0.005 (0.006) |
−0.003 (0.010) |
−0.0002 (0.013) |
|
Total facility |
||||||||
|
RN |
−2.330 (2.562) |
0.836 (2.562) |
−4.771 (3.918) |
3.639 (5.096) |
−3.720 (2.933) |
0.743 (3.058) |
−6.550 (5.388) |
5.637 (6.212) |
|
LPN |
−0.443 (0.653) |
−0.258 (0.542) |
−0.771 (1.303) |
−1.227 (1.423) |
0.482 (0.701) |
0.168 (0.738) |
0.358 (1.457) |
−0.080 (1.566) |
|
Area-year fixed effects |
No |
No |
No |
No |
||||
|
Hospital fixed effects |
Yes |
Yes |
Yes |
Yes |
||||
|
Year fixed effects |
Yes |
Yes |
Yes |
Yes |
||||
|
Observations |
9482 |
8878 |
6424 |
12937 |
12637 |
5142 |
||
|
Number of hospitals |
1753 |
1669 |
3212 |
2551 |
2511 |
2571 |
Notes: Separate regressions were run for reclassifiers and declassifiers. All regressions control for Medicare case mix and for teaching, hi-tech and for-profit status. In addition, the RN and LPN regressions also control for hospital unit and nursing home beds, admissions and inpatient days; and outpatient and missing outpatient visits. Medicare share has missing observations. Medicare share observations and number of hospitals, respectively by column: (1) 8960, 1728; (2) 8364, 1646; (3) and (4) 6095, 3186; (5) 12285, 2515; (6) 11984, 2477; (7) and (8) 4904, 2546.
*
Statistically significant difference (p < 0.05); + indicates a statistically significant difference (0.05 ≤ p < 0.10); standard errors were calculated under the assumption that observations were not independent within hospital.
Table A3.
Large Sample of Hospitals: Estimates of the Effect of Geographical Reclassification on Patient Outcomes
|
Patient outcome |
Controls chosen from originating area
|
Controls chosen from destination area
|
||||
|
Ages 40–64 |
Reclassified |
Declassified |
(N) |
Reclassified |
Declassified |
(N) |
|
Mortality |
0.003 (0.002) |
0.005 (0.003) |
294,941 (229,753) |
0.002 (0.002) |
0.0003 (0.002) |
849,408 (872,588) |
|
Pneumonia |
−0.003 (0.002) |
−0.002 (0.003) |
295,222 (229,871) |
−0.001 (0.002) |
−0.007* (0.001) |
849,429 (872,419) |
|
DVT/PE |
0.0013 (0.0008) |
−0.0005 (0.001) |
295,222 (229,871) |
0.0003 (0.0009) |
0.0003 (0.001) |
849,429 (872,419) |
|
Failure to Rescue |
0.003 (0.014) |
0.004 (0.021) |
15,638 (12,209) |
−0.007 (0.014) |
0.015 (0.015) |
49,157 (50,335) |
|
Urinary Tract Infection |
−0.002 (0.002) |
0.004 (0.003) |
295,222 (229,871) |
−0.007* (0.002) |
0.003 (0.002) |
849,429 (872,419) |
|
Pressure Ulcers |
0.004* (0.001) |
−0.004+ (0.002) |
295,222 (229,871) |
0.004* (0.001) |
−0.004* (0.001) |
849,429 (872,419) |
|
Length of stay (natural log) |
0.006 (0.011) |
−0.026+ (0.014) |
284,697 (221,288) |
0.003 (0.008) |
0.024* (0.008) |
830,042 (850,345) |
|
|
||||||
|
Ages 65–80 |
||||||
|
Mortality |
0.001 (0.002) |
0.0007 (0.003) |
560,530 (436,707) |
−0.0005 (0.002) |
−0.0003 (0.002) |
1383,397 (1412,430) |
|
Pneumonia |
0.002 (0.002) |
−0.003 (0.002) |
561,462 (437,197) |
−0.001 (0.001) |
−0.003+ (0.001) |
1383,925 (1412,582) |
|
DVT/PE |
0.0002 (0.0007) |
−0.0002 (0.0008) |
561,462 (437,197) |
−0.0002 (0.001) |
−0.0004 (0.0007) |
1383,925 (1412,582) |
|
Failure to rescue |
0.012 (0.011) |
0.002 (0.018) |
40,085 (31,148) |
0.002 (0.010) |
0.013 (0.011) |
104,300 (106,040) |
|
Urinary tract infection |
−0.003 (0.002) |
0.004 (0.004) |
561,462 (437,197) |
−0.004* (0.002) |
0.003 (0.002) |
1383,925 (1412,582) |
|
Pressure ulcers |
0.002+ (0.001) |
0.0004 (0.002) |
561,462 (437,197) |
0.000 (0.001) |
−0.002+ (0.001) |
1383,925 (1412,582) |
|
Length of stay (natural log) |
−0.003 (0.006) |
−0.006 (0.014) |
548,650 (426,560) |
0.008 (0.006) |
0.026* (0.006) |
1360,613 (1386,767) |
|
Area-year fixed effects |
Yes |
Yes |
|
Yes |
Yes |
|
|
Hospital fixed effects |
Yes |
Yes |
|
Yes |
Yes |
|
|
Year fixed effects |
No |
No |
|
No |
No |
|
|
Number of hospitals |
463 |
433 |
|
558 |
582 |
|
Notes: See notes to Table 5. Separate regressions were run for reclassified and declassified. In the (N) columns, the top number is the number of observations in the reclassified regressions, and the number in parenthesis is the number of observations in the declassified regressions.
*
Statistically significant difference (p < 0.05); +Statistically significant difference (0.05 ≤ p < 0.10); in the following regressions, standard errors were calculated under the assumption that observations were not independent within hospital: ages 40–64 regressions that use the originating area for controls, and the ages 65–80 declassified regressions that use the originating area for controls.
References
Barro, J., Beaulieu, N., 2003. “Selection and Improvement: Physician Response to Financial Incentives.” National Bureau of Economic Research Working Paper 10017, Cambridge, MA.
M. Chalkley, J.M. Malcomson
Government Purchasing of Health Services
A. Culyer, J. Newhouse (Eds.), Handbook of Health Economics, North Holland, Amsterdam (2000)
L. Chan, T.D. Koepsell, R.A. Deyo, P.C. Esselman, J.K. Haselkorn, J.K. Lowery, W.C. Stolov
The effect of Medicare's Payment System for rehabilitation hospitals on length of stay, charges, and total payments
New England Journal of Medicine, 337 (1997), pp. 978–985
|
|
Committee on Quality of Health Care in America, 2001
Committee on Quality of Health Care in America
Crossing the Quality Chasm: A New Health System for the 21st Century
National Academy Press, Washington, DC (2001)
J. Currie, J. Gruber, M. Fischer
Physician payments and infant mortality: evidence from Medicaid Fee Policy
The American Economic Review, 85 (2) (1995), pp. 106–111
D.M. Cutler
The incidence of adverse medical outcomes under prospective payment
Econometrica, 63 (1) (1995), pp. 29–50
|
|
D.M. Cutler
Cost shifting or cost cutting? The incidence of reductions in Medicare payments
J.M. Poterba (Ed.), Tax Policy and The Economy vol. 12, The MIT Press, Cambridge (1998), pp. 1–27
|
D. Cutler, M. McClellan, J. Newhouse
How does managed care do it?
Rand Journal of Economics, 31 (3) (2000)
H.S. Diamond, E. Goldberg, J.E. Janosky
The effect of full-time faculty hospitalists on the efficiency of care at a community teaching hospital
Annals of Internal Medicine, 129 (1998), pp. 197–203
|
|
Doyle, J., 2005. Health Insurance, Treatment and Outcomes; Using Auto Accidents as Health Shocks.” NBER Working Paper #11099, NBER, Cambridge, MA.
D. Dranove, W.D. White
Medicaid-dependent hospitals and their patients: how have they fared?
Health Services Research, 33 (2 Pt 1) (1998), pp. 163–185
|
R.P. Ellis, T.G. McGuire
Provider behavior under prospective payment: cost sharing and supply
Journal of Health Economics, 5 (1986), pp. 129–152
R.P. Ellis, T.G. McGuire
Hospital response to prospective payment: moral hazard, selection, and practice-style effects
Journal of Health Economics, 15 (1996), pp. 257–277
|
|
Employee Benefits Research Institute, 2004
Employee Benefits Research Institute., 2004. Public Attitudes on the U.S. Health Care System: Findings from the Health Confidence Survey. Issue Brief 275, Employee Benefits Research Institute. URL http://www.ebri.org/hcs/.
Gaynor, M., James R., Lowell, T., 2001. Incentives in HMOs.” National Bureau of Economic Research Working Paper 8522, NBER, Cambridge, MA.
P.J. Gertler, D.M. Waldman
Quality-adjusted cost functions and policy evaluation in the nursing home industry
Journal of Political Economy, 100 (6) (1992), pp. 1232–1256
|
|
J. Glazer, T.G. McGuire
Multiple payers, commonality and free-riding in health care: Medicare and private payers
Journal of Health Economics, 21 (2002), pp. 1049–1069
|
|
|
S. Glied
Managed care
J. Newhouse, A. Culyer (Eds.), Handbook of Health Economics, Chapter 13, Amsterdam: Elsevier (2000), pp. 708–745
D.C. Grabowski
Medicaid reimbursement and the quality of nursing home care
Journal of Health Economics, 20 (2001), pp. 549–569
|
|
Grabowski, D.C., Gruber, J., Angelelli, J.J., 2006. “Nursing Home Quality as a Public Good.” NBER Working Paper No. W12361, NBER, Cambridge.
B. Gray
Do Medicaid physician fees for prenatal services affect birth outcomes?
Journal of Health Economics, 20 (2001), pp. 571–590
|
|
J. Gruber
The effect of competitive pressure on charity: hospital responses to price shopping in California
Journal of Health Economics, 13 (2) (1994), pp. 183–212
J. Gruber, J. Kim, D. Mayzlin
Physician fees and procedure intensity: the case of cesarean delivery
Journal of Health Economics, 18 (4) (1999), pp. 473–490
|
|
S. Guterman
The Balanced Budget Act of 1997: will hospitals take a hit on their PPS margins?
Health Affairs, 17 (1) (1998), pp. 159–166
|
D. Hemenway, A. Killen, S.B. Cashman, C.L. Parks, W.J. Bicknell
Physicians’ responses to financial incentives
New England Journal of Medicine, 322 (1990), pp. 1059–1062
B.J. Hillman, C.A. Joseph, M.R. Mabry, J.H. Sunshine, S.D. Kennedy, M. Noether
Frequency and costs of diagnostic imaging in office practice: a comparison of self-referring and radiologist referring physicians
New England Journal of Medicine, 323 (1990), pp. 1604–1608
|
|
D. Hodgkin, T.G. McGuire
Payment levels and hospital response to prospective payment
Journal of Health Economics, 13 (1994), pp. 1–29
|
|
K.L. Kahn, E.B. Keeler, M.J. Sherwood, W.H. Rogers, D. Draper, S.S. Bentow, L.V. Reinisch, L.V. Rubinstein, J. Kosecoff, R.H. Brook
Comparing outcomes before and after implementation of the DRG-based prospective payment system
JAMA, 264 (1990), pp. 1984–1988
|
|
Kaiser Family Foundation, 2004a
Kaiser Family Foundation, 2004a. Kaiser Health Poll Report. URL http://www.kff.org/healthreport/archive_aug2004 last accessed November 2006.
Kaiser Family Foundation, 2004b
Kaiser Family Foundation, 2004b. Kaiser Health Poll Report. URL http://www.kff.org/healthreport/archive_aug2004 last accessed November 2006.
Kaiser Family Foundation, 2004c
Kaiser Family Foundation, 2004c. State fiscal conditions and Medicaid. Kaiser Commission on Medicaid and the Uninsured, Medicaid Facts. URL http://www.kff.org/medicaid last accessed November 2006.
D. Kessler, M. McClellan
Do doctors practice defensive medicine
Quarterly Journal of Economics, 111 (1996), pp. 353–390
|
R.T. Konetzka, D. Yi, E.C. Norton, K.E. Kilpatrick
Effects of Medicare payment changes on nursing home staffing and deficiencies
Health Services Research, 39 (3) (2004), pp. 463–487
J. Kosecoff, K.L. Kahn, W.H. Rogers, E.J. Reinisch, M.J. Sherwood, L.V. Rubinstein, D. Draper, P. Roth, C. Chew, R.H. Brook
Prospective payment system and impairment at discharge; the ‘quicker-and-sicker’ story revisited
JAMA, 264 (1990), pp. 1980–1983
|
Medicare Payment Advisory Commission, 2000. Report to the Congress: Medicare PaymentPolicy, MedPAC, Washington.
M. McClellan
Hospital reimbursement incentives: An empirical analysis
Journal of Economics & Management Strategy, 6 (1) (1997), pp. 91–128
|
|
T. McGuire
Physician agency
J. Newhouse, A. Culyer (Eds.), Handbook of Health Economics, Amsterdam: Elsevier (2000), pp. 462–517
D. Meltzer, W. Manning, J. Morrison, M. Shah, L. Jin, T. Guth, J. Levinson
Effects of physician experience on costs and outcomes on an academic general medicine service: Results of a trial of hospitalists
Annals of Internal Medicine, 137 (2002), pp. 866–874
|
R. Miller, H. Luft
Does managed care lead to better or worse quality of care
Health Affairs, 16 (5) (1997), pp. 7–25
|
|
R. Miller, H. Luft
Does managed care lead to better or worse quality of care
Health Affairs, 16 (5) (2002), pp. 7–25
Needleman, J., Buerhaus, P.I., Mattke, S., Stewart, M., Zelevinsky, K., 2001. Nurse Staffing and Patient Outcomes in Hospitals. Final Report US Department of Health and Human Services, Health Resources and Services Administration Contract No. 230-99-0021.
Needleman, J., Kurtsman, E.T., Kizer, K.W., 2005. Performance Measurement of Nursing Care: State of the Science and Consensus Development. Robert Wood Johnson Foundation, Unpublished paper.
J.P. Newhouse
Pricing The Priceless: A Health Care Conundrum
The MIT Press, Cambridge (2002)
J. Newhouse, D. Byrne
Did Medicare's prospective payment system cause length of stay to fall?
Journal of Health Economics, 7 (4) (1988), pp. 413–416
|
|
|
D. Polsky, S. Nicholson
Why are managed care plans less expensive: risk selection, utilization or reimbursement?
Journal of Risk and Insurance, 71 (2004), pp. 21–40
|
|
T. Rice
The impact of changing Medicare reimbursement rates on physician-induced demand
Medical Care, 21 (1983), pp. 803–815
|
W.D. Rifkin, E. Holmboe, H. Scherer, H. Sierra
Comparison of hospitalists and nonhospitalists in inpatient length of stay adjusting for patient and physician characteristics
Journal of General Internal Medicine, 19 (2004), pp. 1127–1132
|
|
W.H. Rogers, D. Draper, K.L. Kahn, E.B. Keeler, L.V. Rubinstein, J. Kosecoff, R.H. Brook
Quality of care before and after implementation of the DRG-based Prospective Payment System: a summary of effects
JAMA, 264 (1990), pp. 1989–1994
|
R.E. Santerre
The inequity of Medicaid reimbursement in the United States
Applied Health Economics and Health Policy, 1 (1) (2002), pp. 25–32
|
W.J. Scanlon
Medicare Hospital and Physician Payments Geographic Cost Adjustment Important to Preserve Beneficiary Access to Services
Testimony Before the Subcommittee on Health, Committee on Ways and Means, House of Representatives (2002)
Y. Shen
The effect of financial pressure on the quality of care in hospitals
Journal of Health Economics, 22 (2003), pp. 243–269
|
|
|
F.A. Sloan, G.A. Picone, D.H. Taylor, S. Chou
Hospital ownership and cost and quality of care: is there a Dime's worth of difference?
Journal of Health Economics, 20 (1) (2001), pp. 1–21
|
|
|
Staiger, D., Gaumer, G., 1995. Price Regulation and Patient Mortality in Hospitals. World Health Organization, 2004.
W.C. Yip
Physician responses to medical fee reductions: changes in the volume of coronary artery bypass graft (CABG) surgeries in the Medicare and private sectors
Journal of Health Economics, 17 (6) (1998), pp. 675–699
|
|
Zuckerman, S., McFeeters, J., Cunningham, P., Nichols, L., 2004. Changes in Medicaid physician fees, 1998–2003: implications for physician participation. The Policy Journal of the Health Sphere.
S. Zuckerman, S. Norton, D. Verilli
Price controls and Medicare: assessing the volume offset assumption
Medical Care Research and Review, 55 (4) (1998), pp. 457–478
Corresponding author at: Department of Economics and Institute of Government and Public Affairs, University of Illinois at Chicago, 815 West Van Buren Street, Suite 525, Chicago, IL 60607, United States.Tel.: +1 312 996 8227; fax: +1 312 996 1404.
In the first 2 weeks after the IOM (Kohn et al. 1999) report on medical errors was published, there were 72 newspaper articles covering the story that explicitly cited the 98,000 figure according to the NewsBank, Inc. index (http://www.newsbank.com/). Even in 2004, the IOM figure of 98,000 deaths due to medical errors continues to be a major influence on public opinion, as there have been 151 newspaper articles citing this figure.
A recent public opinion poll reported that only 32% of Americans believe that managed care companies are doing a good job serving their patients, and the same survey reported that nearly 60% of Americans are worried that their managed care company will put profits before their health (Kaiser Family Foundation, 2004b).
Medicaid payments may also be kept low because providers subsidize Medicaid payments with payments from private payers. This would be the case if there is joint production and common quality (Glazer and McGuire, 2002 and Grabowski et al., 2006).
In 2004, all 50 states either froze or reduced Medicaid payments to at least one group of medical providers in response to budgetary pressures (Kaiser Family Foundation, 2004c).
For example, see the following studies: Rice (1983), Hillman et al. (1990), Hemenway et al. (1990), Kessler and McClellan (1996), Yip (1998), Gruber et al. (1999), McGuire (2000), Gaynor et al. (2001), Santerre (2002) and Barro and Beaulieu (2003).
For studies of the effect of Medicare reimbursement on services see Rice (1983), Newhouse and Byrne (1988), Ellis and McGuire (1996), Chan et al. (1997), Yip (1998), Zuckerman et al. (1998).
There are only two published studies of the effect of Medicaid fees on health; both examine the effect of fees on infant health and both find that lower fees adversely affect infant health (Currie et al., 1995 and Gray, 2001).
One problem with this series of papers is that they are basically before and after analyses and are therefore unable to control for time-varying influences that may be confounding their results. This is likely to have been problematic since as these papers and others (e.g., Hodgkin and McGuire, 1994) demonstrate, there were significant trends in mortality and treatment at this time.
Due to data limitations, estimates of the effect of changes in average reimbursement in Cutler (1995) are specific to Massachusetts. So while Cutler (1995) avoids the weakness of the simple pre- and post-PPS research design by exploiting the fact that Massachusetts implemented PPS later than other states in New England, identification comes from the assumption that unmeasured trends in mortality in Massachusetts were the same as those in other New England States.
The information about reclassification regulation is taken from the testimony before Congress of William Scanlon, Director of Health Care Issues for the General Accounting Office (Scanlon, 2002).
Note that this limits the size of the increase to between 8% (6%) and 16% (18%). In our data, the average change in the operating wage index for hospitals changing classification is approximately 10%.
For hospitals that “switch” (or are “controls”) more than once, we treat that hospital as a separate hospital each time it switches (or is used as a control).
Short-term, general medical and surgical hospitals make up approximately half of the hospitals in the AHA, and missing AHA data further limited the sample.
In 1994, the following states participated: AZ, CA CO, CT, FL, IL, IA, MD, MA, NJ, NY, OR, PA, WA, and WI. In 1999, participating states were AZ, CA CO, CT, FL, IL, IA, MD, MA, ME, MO, NJ, NY, OR, PA, UT, VA, WA, and WI.
Prior to 1998, the NIS sample was chosen to preserve a longitudinal component—hospitals previously surveyed had a greater likelihood of being surveyed in the following year. This sampling procedure was discontinued in 1998.
The ICD-9 codes for each of these categories is: Septicemia (all ICD-9 beginning with 038); volume depletion (ICD-9 2765); ischemic heart disease (all ICD-9 codes beginning with 410-414); diseases of pulmonary circulation (all ICD-9 codes beginning with 415-417); other forms of heart disease (all ICD-9 codes beginning with 420-429); cerebrovascular disease (all ICD-9 codes beginning with 430-438); diseases of arteries, arterioles, and capillaries (all ICD-9 codes beginning with 440-448); diseases of veins, and lymphatics, and other diseases of the circulatory system (all ICD-9 codes beginning with 451-459); pneumonia and influenza (all ICD-9 codes beginning with 480-487); pneumoconioses and other lung diseases due to external agents (all ICD-9 codes beginning with 500-508); and other diseases of the respiratory system (all ICD-9 codes beginning with 510-519).
Alternative estimation approaches (e.g., Poisson regression) does not change the precision of estimates nor the inferences drawn from estimates.
Copyright © 2007 Elsevier B.V. All rights reserved.
Copyright © 2015 Elsevier B.V. or its licensors or contributors. ScienceDirect® is a registered trademark of Elsevier B.V.
Cookies are used by this site. To decline or learn more, visit our Cookies page.
Top of Form
Bottom of Form
The relationship between nurse staffing and patient outcomes: A case study
2009, International Journal of Nursing Studies
more
Caroline Shuldham, Claire Parkin, Ashi Firouzi, Michael Roughton, Margaret Lau-Walker
The relationship between nurse staffing and patient outcomes: A case study
International Journal of Nursing Studies, Volume 46, Issue 7, July 2009, Pages 986–992
Original Research Article
Is the impact of managed care on hospital prices decreasing?
2008, Journal of Health Economics
more
David Dranove, Richard Lindrooth, William D. White, Jack Zwanziger
Is the impact of managed care on hospital prices decreasing?
Journal of Health Economics, Volume 27, Issue 2, March 2008, Pages 362–376
Original Research Article
The effect of financial pressure on the quality of care in hospitals
2003, Journal of Health Economics
more
Yu-Chu Shen
The effect of financial pressure on the quality of care in hospitals
Journal of Health Economics, Volume 22, Issue 2, March 2003, Pages 243–269
Original Research Article
This article has not been cited.
No articles found.
Saved to reference managers
1 CiteULike
+ Show more applications
- Show fewer applications
Top of Form
Bottom of Form
Download PDFs
Entitle Gadget