Ownership Status and Length of Stay in Skilled Nursing Facilities

Abstract

In 1998, Medicare implemented the Prospective Payment System for post-acute care provided by skilled nursing facilities. This system paid a fixed price per day above the cost of care, creating an incentive to provide longer length of stays to increase revenues. In this paper, we examine whether there are systematic differences in length of stay for post-acute care patients between for-profit and not-for-profit skilled nursing facilities. Based on the financial incentives inherent in the reimbursement system, we develop a conceptual framework that argues for-profits will provide a greater number of days of care to increase profits relative to not-for-profits. We find significant differences in length of stay by ownership, but once patient selection into a facility is accounted for using two-staged residual inclusion, there is no statistical differences in length of stay between for-profit and not-for-profit facilities.

Keywords

skilled nursing facilities post-acute care Medicare length of stay reimbursement endogeneity

Introduction

Policy makers are interested in creating reimbursement systems that incentivize providers to control costs while maintaining quality. One way to do this is to institute reimbursement systems that pay providers a fixed price and permit them to keep any cost savings (Newhouse, 1996, 2004). One such system is Medicare’s Prospective Payment System (PPS). For hospitals, PPS pays a fixed price per hospitalization. After hospitals implemented the PPS, patients’ length of stay (LOS) decreased. This decreased LOS helped hospitals reduce the cost of care, but may have harmed some patients because they were discharged too soon (Ellis & McGuire, 1996; Freiman, Ellis, & McGuire, 1989; Newhouse & Byrne, 1988).

Although there is an extensive literature related to the effect of the PPS in hospitals, less is known about how the PPS affects post-acute care (i.e., rehabilitation provided after a hospitalization) in skilled nursing facilities (SNFs). At first, the introduction of PPS in 1998 caused significant financial strain, resulting in 5 of the nation’s 10 largest SNF chains declaring bankruptcy (Kitchener, O’Neill, & Harrington, 2005; Qaseem, Weech-Maldonado, & Mkanta, 2007). Subsequent reforms changed reimbursement rates and, over time, SNFs learned to develop strategies to increase revenue. This caused Medicare expenditures on SNF-based post-acute care to increase from $11.2 to $25.5 billion between 1998 and 2009, making post-acute care the fastest growing expenditure component of Medicare during this period (Centers for Medicare & Medicaid Services [CMS], 2010a).

Unlike hospitals, which have an incentive to reduce LOS because they are paid a fixed price per hospital stay, SNFs are paid a fixed price per day that is generally higher than the cost of care (CMS, 2010a). Furthermore, unlike acute care, treatment for post-acute care does not have standard treatment guidelines. As post-acute care is less verifiable and the price of care is generally higher than costs, reimbursement systems that pay on a per diem basis create an incentive for SNFs to increase patient LOS to generate more profits.

Based on the financial incentives inherent in the PPS for post-acute care, we develop a conceptual framework that argues that for-profit (FP) SNFs will provide longer LOS to increase revenues. We then empirically examine whether there are systematic differences between FP and not-for-profit (NFP) LOS for new Medicare-reimbursed post-acute care patients. Although the vast majority of existing studies comparing FP and NFP behaviors treat ownership status as exogenous, a growing literature has found evidence that patient facility selection can lead to endogeneity of ownership status, especially in post-acute care (Bowblis & McHone, 2013; Grabowski, Feng, & Hirth, 2013). This can occur because FP SNFs attempt to attract more profitable patients or because patients with more severe illnesses prefer either FP or NFP SNFs. These unobserved behavioral elements create endogeneity between outcomes such as LOS and ownership status, which can lead to false conclusions about causality, direction of effects, and/or insignificant results. Controlling for this endogeneity through an instrumental variables approach, this study finds little evidence that FPs provided greater lengths of stay even though treating ownership as exogenous suggest shorter LOS in NFP. These results illustrate the dangers of reaching erroneous conclusions when the ownership of SNFs is assumed to be exogenous.

Method

Conceptual Framework

In the optimal reimbursement system, provider incentives are perfectly aligned with the objectives of a regulator, but designing such a reimbursement scheme is difficult. When services are not easily verifiable, providers are likely to adjust their behavior to increase profits (Chalkley & Malcomson, 1998; McGuire, 2000). Furthermore, the longer a reimbursement system is in place, the more time firms have to learn how the system works and can implement strategies to increase profits.

FPs and NFPs are likely to respond differently to the financial incentives from the reimbursement system (Aaronson, Zinn, & Rosko, 1994; Bowblis, 2011; Grabowski et al., 2013; Weisbrod, 1998). Studies have found that NFPs have lower profit margins than FPs, and although there is no clear consensus on the objectives of NFPs, NFPs are thought to have objectives that make them less responsive to financial incentives (see Lindrooth & Weisbrod, 2007; Wachterman, Marcantonio, & Davis, 2011; Weech-Maldonado, Laberge, & Pradhan, 2012). For example, Alchian and Demsetz (1972) and Hansmann (1980) argue that because NFPs cannot redistribute profits, NFPs have less incentive to make profit maximizing decisions. Goodderis and Weisbrod (1998) argue that NFPs may be less efficient than FPs because NFPs can cover operating losses with access to tax-deductible donations and tax exemptions. Dees (1998) notes that NFPs may try to maximize not only profits but also unprofitable “mission” goods such as caring for unprofitable patients. In all of these situations, both FP and NFP would seek the profitable patients, but NFP firms would also seek patients who are less profitable.

To illustrate these differences in incentives, we turn to the hospice industry. Medicare pays hospices a fixed per diem rate. Also, hospices have a U-shaped cost curve where the daily cost of care is higher when the patient is admitted to the hospice and again immediately prior to death. This implies that if a hospice can take advantage of the more profitable intermediate days between admission and death, they can increase profits. This creates an incentive to select patients who require less resource intensive care and have longer LOS. Consistent with FPs being more profit oriented, Lindrooth and Weisbrod (2007) and Wachterman et al. (2011) both find that FP hospices are significantly more likely than NFPs to admit more profitable patients with longer expected LOS.

For SNFs, Medicare PPS for post-acute care creates an incentive to increase LOS. After 3 days of hospitalization, Medicare patients may receive post-acute rehabilitative services in an inpatient setting at a SNF. The cost of providing care depends on local labor market wages, the amount of therapy needed by the patient, and the patient’s illness severity. Medicare reimburses SNFs a fixed amount per day adjusted for these cost factors. These payments are generally greater than the cost of providing care (CMS, 2010a). Thus, profits are larger with longer LOS.

Because there is little clinical consensus on how intensive post-acute care should be, a reimbursement system like PPS creates a financial incentive to increase LOS to increase profits. To reduce the incentive to extend LOS beyond what is clinically necessary, Medicare places limits on the number of days it will fully reimburse care. For the first 20 days of a stay, there is no copayment. From Days 21 to 100, the patient must pay a copayment, which was $114 in 2004 and will be $157.50 in 2015. After 100 days, the patient is fully responsible for the cost of care. However, the vast majority Medicare-eligible patients have supplemental insurance plans that will cover some or all of the cost of the copayment (Gold, Jacobson, Damico, & Neuman, 2013, Exhibit 1). Therefore, if NFPs have weaker profit incentives than FPs, it is expected that LOS would be longer in FPs.

Data and Empirical Models

The primary source of data in the study is the Minimum Data Set (MDS) for calendar year 2005. MDS is a federally mandated assessment for all nursing home residents, including post-acute patients in SNFs. Data are collected by nursing home staff on admission, discharge, and other points during a stay. The assessments contain a variety of information including demographics, diagnoses, clinical items on physical and cognitive functioning, and select payment information (Hawes, Morris, & Phillips, 1995; Morris, Hawes, & Fries, 1990). For the analysis, the MDS is used to identify patients eligible for the study, and to construct the dependent variable of LOS and control variables.

To identify the ownership status of each SNF and obtain other facility-level covariates, MDS is merged with the Online Survey Certification and Reporting (OSCAR) system for all 50 states and the District of Columbia. Data in OSCAR are collected as part of the annual CMS inspection process required of all Medicare and Medicaid certified nursing homes. Although these inspections can be more than 15 months apart, on average, these inspections occur every 12 months. Data collected include ownership, facility structure, and staffing characteristics (CMS, 2010b). OSCAR is supplemented with data from the Area Resource File (ARF) and Rural–Urban Commuting Area (RUCA) codes to obtain local area covariates.

To determine if ownership status affects LOS, only patients admitted to freestanding FP and NFP SNFs are included in the analysis. Hospital-based SNFs are excluded because they may have different case-mixes and LOS compared with freestanding SNFs (CMS, 2007, Table 6.7). MDS data are used to identify all patients 65 and older who are admitted to a SNF in calendar year 2005 for post-acute care paid for by traditional Medicare (i.e., fee-for-service). Additional restrictions are placed on the sample to reduce any heterogeneity in SNF patients that may be correlated with ownership status.

We first restrict the sample to patients who are in the community and experienced their first post-acute care admission. To identify this group, the sample includes patients by their first non-readmission in the calendar year that is directly after an acute care hospitalization, MDS did not indicate a prior history in an institutional setting in the last 5 years (i.e., nursing home, assisted living, group home, or psychiatric setting), and the patient’s LOS is at least 1 day for standard rehabilitation. While this reduces the number of facilities that have at least one admission, these restrictions result in a more homogeneous sample.

In addition, LOS is likely to be a function of the reason the patient is receiving post-acute care. This can be handled by either including a series of indicator variables for diagnoses or, using a more common practice in the literature, restricting the sample to patients with select diagnoses (Bowblis & Brunt, 2014; Wodchis et al., 2005). For completeness, results are reported for three samples: all SNF patients regardless of diagnosis, stoke patients, and hip fracture patients. Stroke and hip fracture patients are included because they are the most common reason for post-acute care rehabilitation in SNFs.

Dependent variable

The dependent variable is the Medicare-reimbursed post-acute care LOS. As MDS does not directly report LOS, it must be calculated from the admission and discharge assessments. The naïve approach is to calculate the LOS as the difference between the admission and discharge dates. However, this approach ignores important institutional details related to post-acute care. First, a copayment is required starting on the 21st day, and Medicare will only cover the first 100 days of post-acute care. Second, patients who are admitted for post-acute care may not show improvement in their functional status and are converted to long-term care patients. Medicare does not cover long-term care and MDS does not contain an assessment for when this conversion occurs. Therefore, defining LOS as the difference between admission and discharge assessment dates can over-estimate the LOS for Medicare-reimbursed post-acute care.

To address these issues, a three-step approach is used. First, the naïve approach is used to determine LOS. Second, any patient with LOS greater than 100 days is reclassified as having a LOS of 100 days and an additional variable is created that identifies these patients as censored. Third, the sample is restricted to those patients for whom the entire LOS is likely to be paid for by Medicare. In particular, Medicare post-acute care patients are required to have assessments approximately 5, 14, 30, 60, and 90 days after admission. If a patient has a Medicare 90-day assessment, then the patient’s entire LOS, up to 100 days, is likely to be paid for by Medicare.

The Medicare-reimbursed LOS is based on the following criteria: (a) LOS of 100 days if the patient has a Medicare 90-day assessment, (b) up to 90 days if the patient has a Medicare 60-day assessment, (c) up to 60 days if the patient has a Medicare 30-day assessment, and (d) up to 30 days if the patient has a Medicare 14-day assessment. Using these criteria, 8.4% of patients could not be identified as having a LOS paid entirely by Medicare. Following these criteria resulted in a sample size of 241,990. More restrictive alternative cut-off criteria (e.g., up to 60 or 75 days if the patient had 60-day assessment) were used to check the robustness of the procedure and the relative differences in LOS between FP and NFP SNFs. Results are found to be robust across various definitions.

Key explanatory and control variables

The key explanatory variable is ownership. Ownership is identified from OSCAR, with facilities classified into FPs and NFPs. Of the 241,990 patients in the entire sample, 187,356 were admitted to a FP and 54,634 to NFPs.

Additional explanatory variables are included as controls as they may influence the LOS for post-acute care. These variables are broadly defined as patient demographics, patient case-mix, facility organizational characteristics, facility staffing levels, and market characteristics.

Patient demographics and case-mix controls are included because these factors can affect LOS. Patient demographics could affect the willingness of patients to be admitted for longer periods and patients with more severe medical conditions have longer LOS. Systematic differences in either of these variables across ownership could bias the results if they are not included in the model. All patient demographics and case-mix variables are constructed from the MDS using the initial admission assessment. Demographics include gender, race, region, and educational attainment. Case-mix adjusters include an indicator variable for functional limitation as measured by the activities of daily living (ADL) index score used by CMS to classify patients into reimbursement categories (CMS, 2004). Patient ADL index scores are broken into scores of 4 to 8, 9 to 14, and 15 to 18. Case-mix is also measured using indicator variables for common medical conditions: stroke diagnosis, hip fracture diagnosis, dementia diagnosis, feeding tube use, and catheter use.

Facility organizational and staffing characteristics may also affect LOS. Organizational characteristics include facility size, chain membership, payer mix, and occupancy rates. More clinical resources such as higher nurse staffing levels and higher rehabilitative staffing levels could reduce LOS by providing better quality care or could increase LOS to help pay for these resources. All staffing variables are measured in hours per resident day (HPRD). For each staff type, staffing levels are considered miscoded if the level of staffing is (a) more than 24 hr, (b) 0, or (c) more than 3 SDs from the mean. Following the previous literature (Bowblis, 2011), the HPRD is set to 0 for miscoded staffing levels and indicator variables for miscoded staffing are coded as 1. As a robustness check, regressions that excluded all staffing variables found similar results.

Local area characteristics may also affect LOS. Market variables include urban versus rural indicator variables, the number of SNF beds per 1,000 Medicare beneficiaries, the percentage of population on Medicare, and the number of hospital admissions per 1,000 persons. Supply and demand measures are constructed from ARF and reflect county-level values in 2005. Urban versus rural is defined using RUCA Categorization A, which aggregates zip codes into four distinct urban–rural categories: urban focused, large rural/town (micropolitan) focused, small rural town focused, and isolated small rural town focused (WWAMI Rural Health Research Center, 2010). In addition, state indicator variables are included to control for regional and regulatory differences across the states.

Empirical method

As the dependent variable is the amount of time until discharge and some observations have censored LOS, survival analysis techniques are used. The analytical strategy starts with the presentation of non-parametric summary statistics and tests that compare the LOS by ownership type, such as the median and mean LOS. Graphical evidence is provided and log-rank tests are performed to determine if the survivor functions are different across ownership. The survivor function is the probability of remaining at the SNF (not discharged) at time t given that the beneficiary remained in SNF care through time t. Finally, to formally control for other factors that can influence LOS, Cox proportional hazards regressions are estimated. The hazard function is directly related to the survivor function except that the hazard function reports the probability of discharge at time t given that the beneficiary remained under SNF care till time t. As LOS across individual patients could be correlated within SNFs, all standard errors are adjusted for clustering of patients within facilities.

Endogeneity

One concern with this analysis is that SNF patients may not be randomly assigned across ownership types even with the host of control variables and sample restricted discussed above. If one type of facility disproportionately takes sicker patients, it will have longer LOS regardless of financial incentives. This patient selection will lead to endogeneity and treating ownership status as exogenous in a regression will lead to biased coefficient estimates. To account for this endogeneity, instrumental variables approaches are utilized to estimate the true causal effect of FP status on LOS. Because the Cox proportional hazards model is a non-linear model, we estimate the model that treats ownership as endogenous using two-stage residual inclusion (Terza, Basu, & Rathouz, 2008).

Two-staged residual inclusion requires that two regressions be estimated. In the first regression, a linear regression is estimated with ownership type as the dependent variable. On the right-hand side is the set of control variables described above and an exclusion restriction. In this case, an exclusion restriction is a variable that explains ownership (endogenous variable) but does not explain LOS (dependent variable of interest). From this first regression, we estimate the predicted residuals for each observation. The predicted residuals from this first-stage regression are then included as an additional control variable in the second regression, the Cox proportional hazards model. By including this predicted residual, we can make causal inference on the coefficient estimates of ownership type in the second-stage regression. In addition, if the coefficient estimates for the predicted residuals are found to be statistically significant, this is evidence that ownership type is endogenous.

The strength of any instrumental variables approach is the validity of the exclusion restriction. The exclusion restriction must explain the endogenous variable, but not affect the dependent variable of interest except through the endogenous variable. For this application, the exclusion restriction is the differential distance between the patient’s home and the closest NFP and FP SNFs. More specifically, the distance between the patient’s home and the closest NFP and FP SNF are calculated. Differential distance is defined as the distance to the closest NFP minus the distance to the closest FP. Negative values of differential distance indicate that a NFP is closer whereas positive values indicate that a FP is closer.

Differential distance is correlated with the decision to use a NFP SNF, and has no theoretical impact on the LOS making it a sound exclusion restriction. Differential distance to a provider is found to be an important determinant of where a patient receives care, and has been used extensively as an exclusion restriction in the health care literature (Gowrisankaran & Town, 1999; McClellan, McNeil, & Newhouse, 1994; Newhouse & McClellan, 1998) and more recently, to study post-acute care quality (Bowblis & McHone, 2013; Grabowski et al., 2013). In the case of SNF post-acute care, Grabowski et al. (2013) provides a strong argument for the validity of differential distance between the FP and NFP SNFs as an exclusion restriction. In our regression context, the differential distance is found to pass standard econometric tests for validity of exclusion restrictions.

Results

Table 1 reports summary statistics for the entire sample and by ownership status. Approximately 23% of our observations occur within NFP SNFs with a differential distance indicating that the average patient is 3.2 miles closer to a FP SNF compared with a NFP SNF. The average beneficiary in our sample is a White female of approximately 81 years of age. The vast majority of patients (55%) have an ADL index score of between 9 and 14. With respect to diagnosis, 21% of patients had dementia, 15% had a stroke, and 10% a hip fracture. The average beneficiary is staying at an urban chain–affiliated SNF with 140 beds.

Table 1.

Summary Statistics.

	All	For-profit	Not-for-profit
Not-for-profit	0.226
	(0.418)
Differential distance	3.205	4.405	−0.908
	(8.687)	(8.917)	(6.302)
Patient demographics
Female gender	0.652	0.646	0.673
	(0.476)	(0.478)	(0.469)
Age	80.912	80.720	81.572
	(7.499)	(7.534)	(7.337)
African American	0.080	0.088	0.053
	(0.272)	(0.284)	(0.224)
Hispanic	0.024	0.027	0.016
	(0.155)	(0.162)	(0.124)
Other/missing race	0.016	0.017	0.012
	(0.125)	(0.130)	(0.107)
Education: <High school (ref. = high school)	0.262	0.271	0.230
	(0.440)	(0.444)	(0.421)
Education: >High school	0.240	0.230	0.272
	(0.427)	(0.421)	(0.445)
Education: Missing	0.018	0.018	0.019
	(0.133)	(0.132)	(0.138)
Patient case-mix
ADL index score: 4-8 (ref. = 15-18)	0.171	0.160	0.209
	(0.377)	(0.367)	(0.406)
ADL index score: 9-14	0.550	0.550	0.549
	(0.498)	(0.497)	(0.498)
Stroke diagnosis	0.148	0.153	0.130
	(0.355)	(0.360)	(0.336)
Hip fracture diagnosis	0.100	0.100	0.103
	(0.301)	(0.300)	(0.304)
Dementia diagnosis	0.208	0.217	0.178
	(0.406)	(0.412)	(0.383)
Feeding tube use	0.040	0.043	0.031
	(0.197)	(0.203)	(0.174)
Catheter use	0.210	0.207	0.222
	(0.408)	(0.405)	(0.416)
Facility characteristics
No. of beds	139.740	134.858	156.484
	(66.669)	(54.136)	(96.305)
Chain membership	0.658	0.718	0.452
	(0.474)	(0.450)	(0.498)
% Patients on Medicaid	53.141	54.253	49.325
	(21.725)	(21.687)	(21.422)
% Patients on Medicare	22.092	22.943	19.173
	(15.102)	(14.770)	(15.848)
Occupancy rate	86.223	86.154	86.458
	(14.391)	(13.874)	(16.036)
Facility staffing
Registered nurse (HPRD)	0.359	0.332	0.449
	(0.308)	(0.274)	(0.391)
Licensed practical nurse (HPRD)	0.792	0.795	0.782
	(0.359)	(0.350)	(0.390)
Certified nurse aids (HPRD)	2.142	2.092	2.310
	(0.690)	(0.685)	(0.684)
Occupational therapy (HPRD)	0.157	0.161	0.144
	(0.141)	(0.135)	(0.158)
Physical therapy (HPRD)	0.206	0.209	0.197
	(0.179)	(0.174)	(0.195)
Market characteristics
Urbanicity: Micropolitan (ref. = urban)	0.111	0.111	0.110
	(0.314)	(0.315)	(0.313)
Urbanicity: Small rural town	0.056	0.059	0.043
	(0.230)	(0.236)	(0.204)
Urbanicity: Isolated rural town	0.026	0.024	0.033
	(0.158)	(0.152)	(0.178)
SNF beds per 1,000 Medicare beneficiary	40.412	39.530	43.435
	(15.798)	(15.900)	(15.055)
% Population on Medicare	15.861	15.830	15.966
	(4.680)	(4.700)	(4.611)
Hospital admission per 1,000 persons	136.634	133.123	148.672
	(82.740)	(79.626)	(91.612)
Observations	241,990	187,356	54,634

Note. Means and standard deviations are reported in the table. Results are not reported for state indicator variables. ADL = activities of daily living; HPRD = hours per resident day; SNF = skilled nursing facility.

Table 2 reports the distribution of the LOS by ownership. The median LOS for the entire sample is 22 days, which is close to the number of days Medicare fully covers post-acute care before a copayment is required. The overall mean LOS is 29.1 days. This number is slightly longer than the 26.9 day LOS reported for all beneficiaries in 2005 that includes admissions that are not considered new episodes (CMS, 2007; Table 6.7). FPs have longer average and median LOS (30.0 and 22 days, respectively) than NFPs (26.1 and 20 days, respectively). This pattern is made even more evident through an examination of the distribution of LOS by ownership exhibited in Figure 1 for the entire sample. NFPs have a substantially larger proportion of patients with LOS below 20 days across each of these samples. FPs, on the other hand, have a larger proportion of patients with LOS exceeding 30 days. This pattern remains evident through LOS of 100 days.

Table 2.

Length of Stay in Days by Ownership.

	All	For-profit	Not-for-profit
All SNF patients
No. of patients	241,990	187,356	54,634
% of patients	100.0	77.4	22.6
Median length of stay	22.00	22.00	20.00
Mean length of stay	29.12	30.00	26.10
Stroke sample
No. of patients	35,815	28,706	7,106
% of patients	100.0	80.2	19.8
Median length of stay	25.00	25.00	23.00
Mean length of stay	33.56	34.10	31.40
Hip fracture sample
No. of patients	24,318	18,697	5,621
% of patients	100.0	76.9	23.1
Median length of stay	28.00	28.00	26.00
Mean length of stay	34.42	35.22	31.73

Note. The mean length of stay refers to restricted mean. SNF = skilled nursing facility.

Figure 1.

Distribution of length of stay: Entire sample length of stay by ownership.

To visually represent the LOS for the entire sample, Figure 2 presents the Kaplan–Meier estimate of the survivor function by ownership type. For all facilities, when LOS is near zero, no patients are discharged and the survival rate estimate is 100%. As the LOS increases, the percentage of patients still in the SNFs declines rapidly, with an inflection point around 20 days. After the inflection point, the percentage of patients still in the facility declines at decreasing rates. Although the general pattern is similar for each ownership type, NFPs have lower survivor functions than FPs, suggesting shorter LOS for patients in NFPs. Log-rank tests of the survivor functions for each sample confirm that LOS is different across ownership (p < .001 for all samples). All of the graphical evidence and univariate statistical tests point to shorter LOS in NFPs.

Figure 2.

Kaplan–Meier survival estimates: Entire sample length of stay by ownership.

The results of the Cox proportional hazards regressions are reported in Table 3. The table reports results in terms of hazard ratios and treats ownership as exogenous in columns labeled (1) and endogenous in columns labeled (2). A hazard ratio of 1 implies no difference in the hazard function as the covariate changes. A hazard ratio greater than 1 is consistent with a higher risk of discharge. In contrast, a hazard ratio less than 1 implies a lower risk of discharge. The LOS is expected to be shorter the greater the risk of discharge.

Table 3.

Cox Proportional Hazards Regression Results for the Effect of Ownership on the Probability of Discharge.

	Entire sample		Stroke sample		Hip fracture sample
	(1)	(2)	(1)	(2)	(1)	(2)
NFP facility (ref. = FP facility)	1.124***	0.949	1.066***	0.966	1.128***	0.967
NFP facility (ref. = FP facility)	(0.014)	(0.054)	(0.020)	(0.109)	(0.025)	(0.107)
First-stage residual		1.191***		1.107		1.174
First-stage residual		(0.070)		(0.127)		(0.133)
First-stage statistics
Exclusion restriction: Differential distance		−0.009***		−0.008***		−0.009***
Exclusion restriction: Differential distance		(0.001)		(0.001)		(0.001)
Staiger–Stock F test on exclusion restrictions		248.20		193.47		176.42
Observations	241,990	241,990	35,815	35,815	24,318	24,318

Note. Hazard ratios and standard errors robust to cluster within facility are reported. The first column treats ownership as exogenous. The second column treats ownership as endogenous using two-staged residual inclusion. Regressions control for patient characteristics, facility structure, staffing, local area characteristics, and state indicator variables.

p < .1. **p < .05. ***p < .01.

For the entire sample and patients with each diagnosis, patients in FPs have the lowest risk of discharge when ownership is treated as exogenous. The hazard ratio of 1.12 for NFPs in the entire sample implies that the risk of discharge is about 12.0% higher in NFP facilities. The corresponding percent increase in risk of discharge in NFPs is 6.6% and 12.8% for stroke and hip fracture patients, respectively.

As shown in Table 1, FPs have a greater proportion of patients in the highest ADL index score range of 15 to 18, and more patients that have an active diagnosis of a stroke or dementia, and are more likely to use feeding tubes. This suggests that FPs may have sicker patients, which could increase the average LOS in FPs, implying ownership is endogenous. Using two-staged residual inclusion to address endogeneity, the hazard ratio for NFP ownership becomes below 1 and is no longer statistically significant in all three samples. This implies that NFPs and FPs have similar average LOS. The first-stage residual is found to be statistically significant, confirming that ownership is endogenous. Furthermore, the exclusion restriction, which is differential distance from the patients’ home and the closest NFP and closest FP, finds that patients are less likely to go to a NFP if the SNF is further away from their home. Also, the Staiger–Stock test finds that the exclusion restriction is not weak. These findings are consistent with validity of the exclusion restriction.

Conclusion

This study compares the average LOS in FP and NFP SNFs. Given the structure of how Medicare reimburses for post-acute care, it is expected that FPs would have longer LOS. Using Cox proportional hazard models, we find that after adjusting for patient and facility characteristics, patients in FP SNFs have longer average and median LOS than patients in NFPs. However, once the endogeneity of ownership status and patient selection is taken into account, we find no significant difference in the LOS of FP and NFP SNFs. Without controlling for the endogeneity of ownership status, one might be led to conclude that either FPs are increasing LOS to increase revenue from Medicare or FPs provide higher quality of care. From a policy perspective, not properly accounting for the endogeneity of ownership would wrongly suggest that FPs are using ambiguities in Medicare post-acute care treatment guidelines to increase revenues.

Moving forward, the Affordable Care Act is expected to bundle SNF and hospital payments. Even though how payments are to be bundled is currently unknown, the results of this article still have importance. FPs and NFPs clearly admit different types of patients, and this patient selection makes it much harder to appropriately compare outcomes between different ownership types or other facility characteristics. Bundled payments only compound this issue as selection may occur between hospitals and SNFs. Even if we know how bundled payments will work, patient selection will always be an issue, and the results of this article show that blindly comparing FPs and NFPs to examine associations between ownership type without accounting for patient selection can be misleading.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Author Biographies

John R. Bowblis, PhD, is an associate professor of economics and research fellow with the Scripps Gerontology Center at Miami University. His research primarily focuses on health economics and health policy with an emphasis on how economic incentives created by regulation and reimbursement systems affect health care provider behavior.

John Horowitz, PhD, is a professor and chair of the economics department at Ball State University. His research focuses on public finance and health economics with an emphasis on how incentives affect providers and buyers.

Christopher S. Brunt, PhD, is an assistant professor of economics at Georgia Southern University. His research focuses on health economics with a predominant emphasis on provider incentives and payment under Medicare Part B, skilled nursing facilities, and the industrial organization of health care.

References

Aaronson

W. E.

Zinn

J. S.

Rosko

M. D.

(1994). Do for-profit and not-for-profit nursing homes behave differently? The Gerontologist, 34, 775-786.

Alchian

A. A.

Demsetz

(1972). Production, information costs, and economic organization. American Economic Review, 62, 777-795.

Bowblis

J. R.

(2011). Ownership conversion and closure in the nursing home industry. Health Economics, 20, 631-644.

Bowblis

J. R.

Brunt

C. S.

(2014). Medicare skilled nursing facilities and upcoding. Health Economics, 23, 821-840.

Bowblis

J. R.

McHone

H. S.

(2013). An instrumental variables approach to post-acute care nursing home quality: Is there a dime’s worth of evidence that continuing care retirement communities provide higher quality? Journal of Health Economics, 32, 980-996.

Centers for Medicare & Medicaid Services. (2004). Medicare program; Prospective payment system and consolidated billing for skilled nursing facilities-update-notice (Vol. 69 FR 45775). Retrieved from http://www.gpo.gov/fdsys/pkg/FR-2004-07-30/pdf/04-17443.pdf

Centers for Medicare & Medicaid Services. (2007). Medicare & Medicaid statistical supplement, 2007 edition. Retrieved from https://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-Reports/MedicareMedicaidStatSupp/2007.html

Centers for Medicare & Medicaid Services. (2010a). Medicare & Medicaid statistical supplement. Retrieved from https://www.cms.gov/MedicareMedicaidStatSupp/09_2010.asp#TopOfPage

Centers for Medicare & Medicaid Services. (2010b). States operations manual. Baltimore, MD. Retrieved from http://www.cms.gov/Regulations-and-Guidance/Guidance/Manuals/Downloads/som107c07.pdf

10.

Chalkley

Malcomson

J. M.

(1998). Contracting for health services when patient demand does not reflect quality. Journal of Health Economics, 17, 1-19.

11.

Dees

J. G.

(1998). Enterprising non-profits. Harvard Business Review, 76, 54-69.

12.

Ellis

R. P.

McGuire

T. G.

(1996). Hospital response to prospective reimbursement: Cost sharing and supply. Journal of Health Economics, 15, 257-277.

13.

Freiman

M. P.

Ellis

R. P.

McGuire

T. G.

(1989). Provider response to Medicare’s PPS: Reductions in length of stay for psychiatric patients treated in scatter beds. Inquiry, 26, 192-201.

14.

Gold

Jacobson

Damico

Neuman

(2013). Medicare Advantage 2013 spotlight: Enrollment market update. The Henry J. Kaiser Family Foundation. Retrieved from https://kaiserfamilyfoundation.files.wordpress.com/2013/06/8448.pdf

15.

Goodderis

J. H.

Weisbrod

B. A.

(1998). Conversion from nonprofit to for-profit legal status: Why does it happen and should anyone care? Journal of Policy Analysis and Management, 17, 215-233.

16.

Gowrisankaran

Town

R. J.

(1999). Estimating the quality of care in hospitals using instrumental variables. Journal of Health Economics, 18, 747-767.

17.

Grabowski

D. C.

Feng

Hirth

(2013). Effect of nursing home ownership on the quality of post-acute care: An instrumental variable approach. Journal of Health Economics, 32, 12-21.

18.

Hansmann

H. B.

(1980). The role of the nonprofit enterprise. The Yale Law Journal, 89, 832-901.

19.

Hawes

Morris

J. N.

Phillips

C. D.

(1995). Reliability estimates for the Minimum Data Set for nursing home resident assessment and care screening. The Gerontologist, 35, 172-178.

20.

Kitchener

O’Neill

Harrington

(2005). Chain reaction: An exploratory study of nursing home bankruptcy in California. Journal of Aging & Social Policy, 17, 19-35.

21.

Lindrooth

R. C.

Weisbrod

B. A.

(2007). Do religious nonprofit and for-profit organizations respond differently to financial incentives? Journal of Health Economics, 26, 342-357.

22.

McClellan

McNeil

B. J.

Newhouse

J. P.

(1994). Does more intensive treatment of acute myocardial infarction in the elderly reduce mortality? Analysis using instrumental variables. Journal of American Medical Association, 272, 859-866.

23.

McGuire

T. B.

(2000). Physician agency. In Cuyler

A. J.

Newhouse

J. P.

(Eds.), Handbook of health economics (pp. 461-536). Amsterdam, The Netherlands: Elsevier.

24.

Morris

J. N.

Hawes

Fries

B. E.

(1990). Designing the national Resident Assessment Instrument for nursing homes. The Gerontologist, 30, 293-307.

25.

Newhouse

J. P.

(1996). Reimbursing health plans and health providers: selection verse efficiency in production. Journal of Economic Literature 34, 1236–1263.

26.

Newhouse

J. P.

(2004). Reimbursing for health care services. Economie publique/Public Economics, 13, 5-33.

27.

Newhouse

J. P.

Byrne

D. J.

(1988). Did Medicare’s prospective payment system cause length of stay to fall? Journal of Health Economics, 7, 413-416.

28.

Newhouse

J. P.

McClellan

(1998). Econometrics in outcomes research: The use of instrumental variables. Annual Review of Public Health, 19, 17-34.

29.

Qaseem

Weech-Maldonado

Mkanta

(2007). The Balanced Budget Act and the supply of nursing home subacute care. Journal of Health Care Finance, 34, 38-47.

30.

Terza

J. V.

Basu

Rathouz

P. J.

(2008). Two-stage residual inclusion estimation: Addressing endogeneity in health econometric modeling. Journal of Health Economics, 27, 531-543.

31.

Wachterman

M. W.

Marcantonio

E. R.

Davis

R. B.

(2011). Association of hospice agency profit status with patient diagnosis, location of care, and length of stay. Journal of the American Medical Association, 305, 472-479.

32.

Weech-Maldonado

Laberge

Pradhan

(2012). Nursing home financial performance: The role of ownership and chain affiliation. Health Care Management Review, 37, 235-245.

33.

Weisbrod

(1998). Modeling the nonprofit organization as a multi-product firm: A framework for choice. In Weisbrod

(Ed.), To profit or not to profit: The commercial transformation of the nonprofit sector (pp. 47-64). Cambridge, UK: Cambridge University Press.

34.

Wodchis

W. P.

Teare

G. F.

Naglie

Bronskill

S. E.

Gill

S. S.

Hillmer

M. P.

. . . Fries

B. E.

(2005). Skilled nursing facility rehabilitation and discharge to home after stroke. Archives of Physical Medicine and Rehabilitation, 86, 442-448.

35.

WWAMI Rural Health Research Center. (2010). RUCA data: Using RUCA data. Retrieved from http://depts.washington.edu/uwruca/ruca-uses.php