|Home | About | Journals | Submit | Contact Us | Français|
To determine the effect of competition in postacute care (PAC) markets on resource intensity and outcomes of care in inpatient rehabilitation facilities (IRFs) after prospective payment was implemented.
Medicare claims, Provider of Services file, Enrollment file, Area Resource file, Minimum Data Set.
We created an exogenous measure of competition based on patient travel distances and used instrumental variables models to estimate the effect of competition on inpatient rehabilitation costs, length of stay, and death or institutionalization.
A file was constructed linking data for Medicare patients discharged from acute care between 2002 and 2003 and admitted to an IRF with a diagnosis of hip fracture or stroke.
Competition had different effects on treatment intensity and outcomes for hip fracture and stroke patients. In the treatment of hip fracture, competition increased costs and length of stay, while increasing rates of death or institutionalization. In the treatment of stroke, competition decreased costs and length of stay and produced inferior outcomes.
The effects of competition in PAC markets may vary by condition. It is important to study the effects of competition by diagnostic condition and to study the effects across populations that vary in severity. Our finding that higher competition under prospective payment led to worse IRF outcomes raises concerns and calls for additional research.
Every year millions of Medicare beneficiaries are discharged from acute care hospitals into institutional postacute care (PAC) in inpatient rehabilitation facilities (IRFs), skilled nursing facilities (SNFs), and long-term care hospitals (LTCHs). Each of these institutional settings offers a different level of care. IRFs provide the most intensive care rehabilitation care (3 or more hours a day of rehabilitation therapy). SNFs can also provide inpatient rehabilitation under the Medicare benefit, although it is generally less intensive than that provided in IRFs (Gage 1999).
From 1988 to 1997, Medicare expenditures for PAC grew at an average annual rate of 25 percent, making it the fastest-growing category of Medicare spending [Medicare Payment Advisory Commission (MedPAC) 2003]. Congress responded by mandating prospective payment for PAC providers. Between 1998 and 2002, Medicare introduced prospective payment systems (PPSs) for SNFs (1998), HHAs (2000), IRFs (2002), and LTCHs (2002). The IRF PPS uses per discharge payments to provide incentives for limiting costs per rehabilitation stay. Payment amounts are based on patient categories defined by the patient's rehabilitation impairment (e.g., stroke, hip fracture), functional status, and comorbidities (Carter et al. 2002). The new payment systems blunted the rate of growth in Medicare expenditures for PAC, although these expenditures continued to rise steadily and now account for about 11 percent of total Medicare spending (Buntin, Colla, and Escarce 2009; MedPAC 2009;).
Researchers have examined the effect of the new payment systems in PAC on resource use and quality and outcomes of care. Early studies found small decreases in SNF utilization, accompanied by increases in the use of other PAC providers, after implementation of the PPS for SNFs (Angelelli et al. 2002; McCall et al. 2003; MedPAC 2003; Buntin et al. 2009;). Rates of adverse outcomes, including acute care readmissions and mortality, did not change (Angelelli et al. 2002; McCall et al. 2003;). By contrast, a later analysis found worsening of certain SNF outcomes between 2000 and 2004, and other studies have found declines in staffing and the intensity of services (Yip, Wilber, and Myrtle 2002; White 2003, 2005; Wodchis, Fries, and Hirth 2004; Murray et al. 2005; MedPAC 2006). Most recently, a study of 120 IRFs found that these facilities reduced costs per discharge during the first year of the IRF PPS (McCue and Thompson 2006). In a national study, Sood, Buntin, and Escarce (2008) found that implementation of the IRF PPS led to sizable declines in costs and length of stay, especially among facilities that had higher payment limits under the preexisting payment system, but patients' rates of return to the community and mortality were unaffected.
The payment system is not the only economic factor that affects resource use and quality of care among health care providers. An extensive body of research on acute care hospitals has shown that market structure, especially the degree of competition, matters as well. Studies of acute care hospitals, moreover, have found that the effects of competition on costs and quality depend on the way providers are paid and how prices for their services are set (e.g., Robinson and Luft 1985; Zwanziger and Melnick 1988; Keeler, Melnick, and Zwanziger 1999; Kessler and McClellan 2000; Mukamel, Zwanziger, and Tomaszewski 2001; Bundorf et al. 2004; Escarce, Jain, and Rogowski 2006; Rogowski, Jain, and Escarce 2007;). For instance, competition increased costs under cost-based reimbursement, whereas it decreased costs under selective contracting based on prices.
Much less is known about the effects of competition in PAC markets. Further, the available research has focused on long-term care services provided in nursing homes, rather than on PAC. Studies conducted in the 1990s found that higher competition among nursing homes was associated with better structural and process quality and with adoption of total quality management (Zinn 1994; Zinn, Weech, and Brannon 1998;). Similarly, recent studies indicate that higher competition is associated with higher scores on the quality measures reported on the Centers for Medicare and Medicaid Services Nursing Home Compare Website (Zinn et al. 1998; Starkey, Weech-Maldonado, and Mor 2005; Castle, Engberg, and Liu 2007; Castle, Liu, and Engberg 2008; Centers for Medicare and Medicaid Services 2009;). To our knowledge, no study has assessed the impact of competition on the cost and quality of PAC under the recently implemented PPSs.
To begin to address this gap in the literature, the current study examines the effect of competition in institutional PAC markets on resource use and health outcomes for patients who received PAC in IRFs, the most intensive setting for postacute rehabilitation care, during the first 18 months after prospective payment went into effect. The study focuses on patients discharged from an acute care hospital after a stroke or hip fracture, two of the conditions that most often receive institutional PAC. The empirical analyses incorporate two noteworthy innovations. First, we construct a measure of the competition facing each IRF that accounts for the fact that SNFs may be viable substitutes for IRFs in many clinical situations. Second, we use instrumental variables estimation to account for the likely endogeneity of competition.
Our conceptual framework posits that PAC providers maximize profits (or an increasing function of profits). Additionally, we conceptualize decisions to use PAC and which facility to use as joint decisions made by patients, their families, and their physicians, and influenced by discharge planners at the acute care hospital and staff at PAC facilities. Clinicians are expected to consider medical and rehabilitation needs when making their recommendations. However, psychosocial and institutional factors are likely to play a strong role for many patients. In particular, the family's preferences for having the patient close to home so family members can visit, the family's impressions on visits to different facilities, and the experiences and relationships of the discharge planning staff at the acute care hospital may affect which facilities patients use. Hospitals with IRF and/or SNF subproviders may find it easier to place their patients in those facilities, which often reserve beds for the hospital's patients.
Standard economic models suggest that competition results in higher quality under administered price systems such as prospective payment, because quality is the only dimension on which providers can compete (Gaynor 2006). If quality is costly to produce—the usual assumption—competition is expected to lead to higher costs as well. Further, the sizes of these effects depend on the quality elasticity of the demand for each provider's services: The lower the quality elasticity, the smaller the effects of competition on costs and quality. The quality elasticity of provider-level demand, in turn, is expected to depend on whether patients, families, and health care personnel have reliable and valid information on the quality of alternative providers and use the information in making decisions.
However, the standard models overlook two phenomena that may alter these predictions. First, if competing PAC providers focus on dimensions of quality that may attract patients, such as furnishings and other amenities, but fail to improve patient care processes, higher competition could result in higher costs without improving health outcomes. In the extreme where provider-level demand is much more responsive to amenities than to care processes, providers constrained by administered prices could even substitute amenities for care processes, leading to worse health outcomes as competition increases. The available research suggests that patients and their families find it extremely difficult to assess the quality of care of rehabilitation providers (Magasi et al. 2009).
Second, the quality elasticity of provider-level demand may not be constant, but instead it may vary with the level of competition. One possibility is that higher competition in PAC markets makes patients, families, and health care personnel more sensitive to quality differences across facilities, resulting in higher quality elasticity as competition increases. Another possibility, however, is that higher competition raises search costs and makes it harder to obtain useful information on the quality of alternative providers (Pauly and Satterthwaite 1981). Under these circumstances, acute care hospital personnel could default to always recommending the same facilities, irrespective of other factors. The consequence would be lower quality elasticity as competition increases. If this effect is pronounced, providers may maximize profits by reducing both quality and costs as competition increases.
These observations suggest that the effects of competition on resource use and health outcomes in IRFs under the IRF PPS are theoretically ambiguous. Additionally, it seems reasonable to posit that the size and even the direction of these effects may vary across clinical conditions as a result of differences in the quality elasticity of demand and other relevant factors. Hip fracture and stroke patients may differ markedly in their cognitive abilities, health status, and demographic characteristics, each of which may affect the quality elasticity of demand. For example, lower cognitive functioning in stroke patients could increase the burden on patients and families to discern the amenities offered by IRFs or other dimensions of quality. Similarly, the poor health status of stroke patients, compared with hip fracture patients, could increase search costs.1
We analyzed patient-level Medicare data to determine the effects of competition on the costs and outcomes of PAC in IRFs. The study sample included all patients hospitalized for a stroke or hip fracture from January 2002 through June 2003 who lived in a metropolitan area and used an IRF within 30 days of their acute care discharge. The two study conditions accounted for about 37 percent of admissions to IRFs and represent the two largest groups of patients using inpatient rehabilitation (MedPAC 2009).
To construct the study sample, we used Medicare acute care hospital claims for the study period to identify stroke patients as those with a principal diagnosis of intracerebral hemorrhage (diagnosis code 431.xx), occlusion and stenosis of precerebral arteries with infarction (433.x1), occlusion of cerebral arteries with infarction (434.x1), or acute but ill-defined cerebrovascular disease (436.xx). We identified hip fracture patients as those with a principal diagnosis of fractures of the neck of the femur (820.xx). Hip fracture patients whose fractures could be due to bone metastases or who suffered major trauma to a site other than a lower extremity were excluded (Sood et al. 2008). We then linked these patients to Medicare enrollment and claims data for IRFs, SNFs, and HHAs. We also linked the data with the Minimum Data Set (MDS), which contains data on the universe of nursing home stays, to ascertain whether and when each study patient was in a custodial nursing home.
We identified patients who were admitted to an IRF within 30 days of discharge from the acute care hospital to construct the study sample. However, we excluded patients who died within 30 days of hospital discharge because their use of PAC was effectively truncated.2 We also dropped patients who were residents of nursing homes before the acute care hospitalization, because they would not be expected to return to community residence; patients for whom Medicare was not the primary payer for the acute care stay, because we likely lacked complete information on utilization; and patients who enrolled in a Medicare HMO at any time in the 2 months following discharge from acute care. Overall, we excluded about 3 percent of stroke patients and 2 percent of hip fracture patients.
The final analysis sample consisted of 38,346 patients who were initially admitted to an IRF following a hip fracture and 47,434 following a stroke. These patients were admitted to 911 different IRFs.
We analyzed three study outcomes: cost and length-of-stay for the initial IRF stay and residence in the community at 60, 120, and 180 days after acute care hospital discharge. To obtain the costs of the initial IRF stay, we used he claims data to determine the charges incurred in each department within the IRF. Next, we estimated costs by multiplying the charges for each department by the cost-to-charge ratio for the department, obtained from Medicare cost reports, and we summed the departmental costs to obtain total costs. We also used claims data to obtain the length of each patient's initial IRF stay. Details of the cost calculation are available in Carter et al. (2002).
We used Medicare and MDS data to determine the location of each patient at 60, 120, and 180 days after acute care discharge. We identified patients who had died by then or who on that day were in an acute care hospital, PAC facility (IRF, SNF, or LCTH), or custodial nursing home. We classified the remaining patients in the study sample as residing in the community.
The key explanatory variable in our analyses was the competition faced by each IRF. Kessler and McClellan (2000) have shown that competition measures based on actual patient flows are likely to be endogenous in analyses of the effect of competition on quality of care, because actual patient flows are likely to be influenced by unobserved dimensions of quality. For example, higher quality facilities may attract more patients and from longer distances. To address this concern, we followed Kessler and McClellan and constructed a measure of competition for each IRF based on predicted market shares in three steps.
First, we estimated a PAC facility choice model for each study condition that allowed patients to choose an IRF or an SNF, because these may be substitutes for many patients (McFadden 1973).3 The estimation samples for the choice models were larger than the study samples described earlier, because they included both patients admitted to IRFs and patients admitted to SNFs within 30 days of acute care discharge. The choice set for each patient included all IRFs and SNFs within a specified radius centered on a patient's zip code of residence and constructed to capture 95 percent of all PAC admissions.4 (We constructed different radii by facility type, study condition, year, and census division.) We then used conditional logit models to estimate facility choice as a function of facility type (IRF, freestanding SNF, or SNF unit) and distance from the centroid of the patient's residential zip code to the facility. We also included two-way and three-way interactions among facility type, distance, and selected patient demographic and clinical characteristics. We did not include other facility characteristics at this stage (e.g., teaching status) because these variables could be correlated with unobserved quality and could therefore be endogenous. In the facility choice model, distance was a very strong predictor of facility choice; the main effect was always highly significant and ranged in magnitude from −0.16 to −0.19, depending on the condition and year. In joint significance tests of the main effect for distance and the distance interactions, the χ2 statistic ranged from 62,596 to 120,000.
Next, we applied the model estimates for the relevant condition to each patient in the estimation samples to predict the probability that the patient would use each IRF and SNF in the patient's choice set. Let pijk be the predicted probability that patient i living in zip code k is admitted to PAC provider j. We can compute the predicted number of patients admitted to provider j from zip code k, Njk; the predicted number of patients admitted to any PAC provider from zip code k, Zk; and the predicted number of patients admitted to PAC provider j from any zip code, Fj, as follows:
Then is the predicted share of patients in zip code k who go to provider j, whereas is the predicted share of provider j's patients coming from zip code k.
Finally, we calculated our measure of the competition facing PAC provider j based on the predicted Hirschmann–Herfindahl index (HHI) for the provider, , calculated as follows (Kessler and McClellan 2000):
This HHI can be understood as a weighted mean of HHIs for the zip codes from which provider j draws patients (the term in parentheses), where the weights are the predicted shares of provider j's patients coming from each zip code.5 The competition measure used in our analyses was calculated as .
We estimated multivariate regression models, using individual patients as the unit of analysis, to determine the effect of the competition facing IRFs on the three study outcomes. The key independent variable in the models was the measure of competition for the IRF to which each patient was admitted. The covariates in the models included patient age, gender, race, ethnicity, and whether the patient had dual Medicaid eligibility. We also included a broad array of clinical characteristics drawn from the hospital claims. Condition-specific severity measures included indicators for basilar artery infarct; carotid, vertebral, or multiple artery infarct; and hemorrhagic stroke for the stroke population; and indicators for pertrochanteric fracture, total hip replacement, partial hip replacement, and hip revision for hip fracture patients. Comorbidities were primary cancer with poor prognosis, metastatic cancer, chronic pulmonary disease, coronary artery disease, congestive heart failure, peripheral vascular disease, severe chronic liver disease, diabetes mellitus with and without end-organ damage, chronic renal failure, nutritional deficiencies, dementia, and functional impairment. Complications arising during the acute-care hospital stay that could influence PAC outcomes included postoperative pulmonary compromise, postoperative gastrointestinal hemorrhage, cellulitis or decubitus ulcer, septicemia, pneumonia, mechanical complications due to a device, implant, or graft, shock or arrest in the hospital, postoperative acute myocardial infarction (AMI), postoperative cardiac abnormalities other than AMI, procedure-related perforation or laceration, venous thrombosis and pulmonary embolism, acute renal failure, delirium, and dementia. Additional covariates were PAC facility characteristics, including indicators for freestanding IRFs, rural IRFs, teaching IRFs, ownership status, bed size, and census region, as well as the percentage of low-income patients, volume of Medicare patients, and Medicare wage index. Finally, we included population characteristics of the patient's zip code, including percent poor, percent Hispanic, and percent black; indicators for census regions; and a quarterly time trend.
We estimated linear models for cost and length-of-stay and probit models for the initial residence in the community at 60, 120, and 180 days after acute care hospital discharge. As noted earlier, a potential source of endogeneity in our analyses arose in the construction of competition measures from patient flows. We addressed this source by constructing our measure from predicted, rather than actual, flows. However, a second, distinct potential source of endogeneity arises when we attempt to estimate the regression models described in the preceding paragraph, and this source requires a different approach. The problem is that patients, their providers, and their families choose the IRFs to which patients are admitted. As a result, patients who are sicker (or healthier) in unobserved ways may choose facilities based on characteristics that are correlated with competition. This could induce a correlation between the error term in the regression models and the competition measure, a form of endogeneity.6
To address this second potential source of endogeneity, we used instrumental variables (IV) estimation (Angrist and Imbens 1995). Our IV was a zip code-level HHI, as developed by Kessler and McClellan (2000). Specifically, for patients in zip code k, we calculated this HHI as follows:
can be understood as the weighted average of the provider-level , for all providers in zip code k's relevant geographic market, where the weights are the predicted shares of patients in the zip code who use each provider. Because it was based on patients' probabilistic choices of providers, rather than their actual choices, it avoids the second source of endogeneity discussed above (Kessler and McClellan 2000).7
We corrected standard errors for heteroskedasticity and clustering at the zip code level.
Stroke patients cost more for IRFs to treat (mean=U.S.$14,554) and had longer average length of stay (17.0 days) than hip fracture patients (U.S.$11,139 and 13.9 days) (Table 1). About a quarter of stroke patients were institutionalized or had died at 60 days after acute care hospital discharge, compared with just under one-fifth of hip fracture patients. The percentage of patients who were institutionalized or had died increased to 33 and 24 percent for stroke and hip fracture, respectively, at 120 days after hospital discharge, and declined again 27 and 21 percent at 180 days. Hip fracture patients were older and more likely to be female, while stroke patients were more likely to be black and dually eligible for Medicaid. On average, stroke patients had more comorbidities (1.4 conditions) than hip fracture patients (0.9 conditions).
The facility-level competition measure averaged 0.93 among hip fracture patients and 0.91 among stroke patients. The means of the zip code-level competition measure (the IV) were similar to the means of the facility-level competition measure (Table 1).
We also conducted two sets of bivariate analyses. Thus, we assessed the unadjusted relationships between competition, on one hand, and comorbidities and complications, on the other, to determine whether observed dimensions of health status varied systematically with our key independent variable (Figure 1). We found that the mean number of complications increased across facilities as the level of competition facing facilities rose, whereas the mean number of comorbidities fell.
Similarly, we assessed the unadjusted relationships between competition and costs and between competition and institutionalization or death (Figure 2). For both study conditions, the mean cost for the initial IRF stay and the percentage of patients who were institutionalized or dead at 180 days after hospital discharge tended to increase across facilities as the competition facing facilities rose (Figure 1), although the differences in the rate of institutionalization or death for hip fracture were very small.
Higher competition resulted in higher costs and longer length of stay among patients with hip fracture, other things being equal (Table 2). However, higher use of resources did not produce better outcomes. On the contrary, hip fracture patients who were admitted to IRFs that faced higher levels of competition were more likely to be institutionalized or dead at 60, 120, and 180 days after hospital discharge than their counterparts who were admitted to IRFs that faced low levels of competition. (Table 2 shows the results for institutionalization or death at 180 days.)
In striking contrast, higher competition led to lower costs and shorter lengths of stay among stroke patients. The level of competition facing IRFs did not affect the likelihood of being institutionalized or dead at 60 days after hospital discharge for these patients (results not shown), but stroke patients who were admitted to IRFs that faced higher levels of competition were more likely to be institutionalized or dead at 120 and 180 days after hospital discharge.
To make the magnitudes of the competition effects on costs and institutionalization or death easier to understand, Table 3 reports predicted costs and health outcomes for hip fracture and stroke patients at different levels of competition, adjusted for all the covariates in the regression models. At a competition value of 0.86, the 10th percentile of the competition measure in hip fracture patients, the average cost of an IRF stay after a hip fracture would be lower by U.S.$221, or 2 percent (Table 3), compared with the observed mean cost (Table 1). At this level of competition, moreover, the rate of institutionalization or death at 180 days after hospital discharge among hip fracture patients would be lower by about 1.3 percentage points, compared with the mean. Applying these figures to the approximately 25,000 hip fracture patients per year who use IRFs and reside in U.S. metropolitan areas suggests that reducing competition to this level would produce about U.S.$5.5 million in cost savings and prevent 328 institutionalizations/deaths. Conversely, at a competition value of 0.99, the 90th percentile of the competition measure, the average cost of a hip fracture IRF stay would increase by U.S.$168, and the rate of institutionalization or death at 180 days would rise by 1.0 percentage point. Thus, increasing competition to this level would raise costs by approximately U.S.$4.2 million and lead to an additional 249 institutionalizations/deaths at 180 days after hospital discharge.
The effects of competition on costs for stroke patients are opposite in direction and larger in magnitude. At a value of 0.80, the 10th percentile of the competition measure in stroke patients, the average cost of an IRF stay after a stroke would be higher by U.S.$683, or 5 percent (Table 3), compared with the observed mean. Further, at this level of competition the rate of institutionalization or death at 180 days after hospital discharge among stroke patients would be lower by about 1.5 percentage points, compared with the observed mean. Applying these figures to the approximately 33,000 stroke patients annually who use IRFs and reside in U.S. metropolitan areas suggests that reducing competition to this level would increase costs by U.S.$22.5 million and prevent 580 deaths/institutionalizations at 180 days. Conversely, at a competition value of 0.98, the 90th percentile of the competition measure, the average cost of a stroke IRF stay would fall by U.S.$486, or 3.3 percent, and deaths/institutionalizations would increase by 1.3 percentage points at 180 days posthospital discharge. Increasing competition to this level would result in cost savings of U.S.$16 million and increase deaths/institutionalizations by 413 per year.
Regression results for selected covariates, including patient and facility characteristics, are reported in Table 2.
We assessed the sensitivity of our results to alternative specifications of our regression models. First, we estimated the same models as in our main analyses, but using ordinary least squares to estimate the linear models and conventional maximum likelihood to estimate the probit models, rather than IV methods. (Thus, these models treated the facility-level competition measure as exogenous.) Second, we estimated “reduced-form” models where we replaced the facility-level competition measure with the zip code-level measure (i.e., the IV in our main analyses) as the key explanatory variable.
Table 4 shows that in most cases the direction and significance of the effects of competition on costs and outcomes did not change. However, the marginal effects were smaller for the models where we treated facility-level competition as exogenous compared with the results from the IV models. Of note, these results suggest that patients who were healthier in unobserved ways tended to use IRFs that faced more competition.
To our knowledge, this is the first study to assess the effects of competition in PAC markets in the prospective payment era. Our conceptual framework suggested that, under prospective payment, the influence of competition on resource use and health outcomes is theoretically ambiguous and must be determined empirically. The framework also suggested that the effects of competition on costs and health outcomes may differ across clinical conditions as a result of differences in the quality elasticity of demand and other factors. Consistent with these notions, our empirical analyses found that the effects of competition on IRF costs and outcomes were complex and differed between our two study conditions, hip fracture and stroke. Specifically, we found that higher competition increased resource use and led to worse outcomes for hip fracture patients, whereas higher competition reduced resource use and led to worse outcomes among stroke patients.
Our findings suggest that IRFs may be competing for hip fracture patients on costly dimensions of care, but that the extra costs do not improve health outcomes; instead, the extra costs are associated with worse outcomes. One possibility is that facilities are spending more on furnishings and other amenities that may appeal to families and patients but do not influence outcomes, rather than focusing on the therapy needs of patients (Goldman and Romley 2008; Romley and Goldman 2008;). Unfortunately, our data did not enable us to explore the mechanisms underlying the deleterious effects of competition on health outcomes in more detail. A minimum of 3 hours of therapy is mandated daily in IRFs. However, studies have found considerable variation across providers in rehabilitation practices for hip fracture patients, and that these practices influence outcomes (Munin et al. 2005; Chudyk et al. 2009;). Possibly, facilities that produce better hip fracture outcomes are able to provide more therapy than the mandated minimum, or different, more effective, types of therapy without increasing costs, by focusing on care processes. Studying the determinants of rehabilitation outcomes for hip fracture patients is an important area for future research.
By contrast, IRFs that face higher levels of competition appear to scale back on resource use for stroke patients. Higher competition also has a negative effect on stroke outcomes by 120 and 180 days posthospital discharge. Stroke patients or their families may have a lower quality elasticity of demand due to, for example, reduced cognitive functioning that would make it harder for them to discern the amenities offered by IRFs or other dimensions of quality (Heruti et al. 1999; Magasi et al. 2009;). The poor health status of stroke patients, compared with hip fracture patients, might also make it harder for them to search for a facility they like. Interestingly, studies have found that certain components of a rehabilitation program, such as exercise training and practice targeted at specific tasks, are the most important in improving function (Kwakkel 2006; Dewey, Sherry, and Collier 2007;). If IRFs reduce the costs of treating stroke patients by cutting back on these components, they could also compromise outcomes.
Of note, the notion that hip fracture and stroke patients differ in their quality elasticity suggests that either IRFs provide different levels of quality to the two types of patients or the two types of patients tend to be found in different IRFs. In separate analyses, we found some evidence for both phenomena. The correlation between facility-level outcomes for stroke and hip fracture was only 0.27 and the rank correlation was only 0.21. In addition, although the study sample was about evenly split between stroke and hip fracture patients, in one-third of IRFs the ratio of stroke to hip fracture patients exceeded 2-to-1.
Our study has implications for evaluating the effects of payment reform. Our findings highlight that the effects of payment reform on patient outcomes and costs might depend on the competitive environment. For example, there is evidence to suggest that the switch from a cost-based payment system to a PPS is likely to reduce costs on average. However, this study suggests that the magnitude and even the direction of this effect in individual markets might depend on the level of competition. Assessing these effects for different types of providers and different clinical conditions is important for understanding and predicting the effects of payment policy reform.
In evaluating payment policy reform it is also important to consider the effects of the reform on the level of competition in markets. For instance, bundled payments might have the effect of reducing competition if providers contract with fewer PAC facilities and acute care hospitals have more control over which facilities patients use. On the other hand, bundled payments may force postacute providers to compete for contracts from hospitals, introducing some price competition into a current scheme with little to none of it. These changes in competition might have important consequences for patients and should be considered in evaluating bundled payments or other policy reform options being discussed in the current health care debate. Our findings suggest that the effects of bundled payments or other reforms on costs and health outcomes might vary across markets and across patients with different conditions. Evaluations of major payment reforms and regulation and antitrust policies that could make entry, exit, or merging in this market more difficult should take these complexities into account.
Our study has several limitations. First, it is likely that the care provided after the initial IRF stay affects health outcomes. Thus, it may be inappropriate to attribute the competition-related differences in outcomes that we found for hip fracture patients solely to the behavior of IRFs. Studying therapies used after the IRF stay, such as home health care, is an important area for future research (Intrator and Berg 1998; DeJong et al. 2009;). Still, the treatments patients receive within the first few weeks after acute care discharge—that is, in the IRF—are expected to have an important influence on their outcomes.
Second, we could not measure functional gain, a crucial outcome of rehabilitation care, using the available Medicare data. Nonetheless, the ability to return to the community is likely to be a good proxy for functional gain in this case, and an important outcome in its own right for hip fracture and stroke patients. Third, we studied a time period just after the introduction of the PPS for IRFs in January 2002. Consequently, we may have studied a period of transition, where behavior in response to competition under prospective payment had not reached equilibrium. However, this may not be a significant limitation, as facilities could anticipate the change to prospective payment and respond quickly. Prior research has shown that facilities changed resource intensity very quickly in response to incentives introduced by the PPS (Sood et al. 2008).
Empirical evidence is crucial to guiding payment policy and understanding the effects of competition on health care costs and outcomes. This is the first study to document the effects of competition in PAC markets on costs and health outcomes for IRFs after the introduction of prospective payment. Our finding that higher competition under prospective payment for IRFs led to worse raises concerns and suggests a need for additional investigation to confirm the results and clarify the mechanisms. Future research should also examine how competition affects costs and outcomes for patients with other conditions and how payment reforms might alter the effect of competition across markets and patient groups.
Joint Acknowledgment/Disclosure Statement: We would like acknowledge funding from National Institute on Aging (grant number R01AG031260) and Agency for Healthcare Research and Quality (grant number R01HS018541), and programming support from Mark Totten.
1Previous research has demonstrated that IRFs responded to the incentives from prospective payment in the first quarter of implementation, and that the response was similar across conditions (Sood et al. 2008). Therefore, we do not expect that differences between hip fracture and stroke in the transition to prospective payment would affect our results.
2We assessed whether attrition due to death was associated with competition and found that average competition was similar for those who died within 30 days of hospital discharge and those who survived.
3In many instances, referrals to PAC settings are made in the absence of clear clinical criteria that would identify the best setting for maximizing outcomes. Thus, patients and doctors must weigh a range of clinical and nonclinical factors—such as the perceived quality of care delivered by a PAC provider and its convenience—when making these decisions. Researchers examining PAC have observed sizable variations in utilization rates, geographically and by type of discharging hospital, and have documented that the choice of treatment in an IRF or an SNF is substantially influenced by the relative local availability of the two types of facilities (Buntin et al. 2005). All of this suggests substitutability between these institutional settings.
4One concern in estimating facility choice models is that patient choice of facilities might be restricted if acute care hospitals have exclusive relationships with PAC providers. However, we find that acute care hospitals refer patients to many different postacute care facilities and very rarely have exclusive relationships with PAC providers. Approximately 80 percent of providers refer patients to at least five PAC providers or more. Only 2–3 percent of hospitals refer to only one IRF or SNF, so the assumptions in our choice model are consistent with observed referral patterns.
5To examine the precision of the competition measure, we estimate the standard error of our competition measure using the Krinsky–Robb method (Krinsky and Robb 1986, 1990). In particular, we drew a sample of coefficients from the multivariate normal distribution with means of the estimated coefficients and covariance of the covariance matrix. We estimated the competition measures using these coefficients 125 times for each facility and found the mean of the HHI to be 0.0874 and the standard deviation to be 0.0009. Thus, our competition measure based on predicted probabilities is estimated quite precisely.
6Unobserved characteristics of patients may be correlated with facility choice and outcomes if patients use information on outcomes to choose facilities and this information is not fully captured by the observed facility characteristics included in the regressions.
7The correlation between and was r = 0.82 for hip fracture and r = 0.85 for stroke, and the partial F-statistic for the “first-stage” regression of on and the covariates was 258 for hip fracture and 609 for stroke. Despite the high correlations, the two HHI's are not the “same” variable. For most zip codes, more than 100 providers contributed to the zip-code level HHI. The additional exogenous variation that the zip code-level measure offers that the provider-level measure does not offer is that it includes the level of competition that other providers in the market face.
Additional supporting information may be found in the online version of this article:
Appendix SA1: Author Matrix.
Please note: Wiley-Blackwell is not responsible for the content or functionality of any supporting materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.