Trends in the Supply of Inpatient Rehabilitation Facilities Services: 1996 to 2004

Background 

IRFs make up 1 component of what is commonly referred to as the PAC continuum. Other PAC settings include SNFs, LTCHs, and home health agencies. Outpatient clinics, day rehabilitation, and comprehensive outpatient rehabilitation facilities may also be considered part of the continuum. IRFs provide comprehensive, intensive rehabilitation services to patients with physical disabilities.1 IRFs may be freestanding, stand-alone hospitals, admitting patients from a range of acute care hospitals and community settings, or units (IRF units), subproviders located within an acute care hospital.

TEFRA (1982) mandated the implementation of prospective payment for acute care hospitals and prompted the rise of managed care.2 Under prospective payment to acute care hospitals, Medicare paid a predetermined amount per stay based on a patient's case-mix severity, thus transferring much of the financial risk to providers and creating strong incentives for the hospital to reduce costs by lowering LOS.3 Several provider types, including IRFs, were exempt from the acute PPS because their costs were too challenging to estimate accurately.4, 5 Thus, TEFRA identified IRFs as discrete entities for the first time.6 To be certified by Medicare as PPS-exempt, an IRF must, among other requirements, provide preadmission screening and close medical supervision of patients, establish a plan of treatment for all patients, and use a multidisciplinary team approach for all patients.7 The most contested requirement is referred to as “the 75% rule,” by which an established percentage of patients treated in an IRF during a cost-reporting period, “required intensive rehabilitative services for treatment of one or more medical conditions”7 that are included in a list of 13 medical conditions. The actual percentage has varied since implementation of the IRF PPS. It was rarely enforced during the early years of IRF PPS during which this study occurred, and has been enforced at levels of 50% to 60% since 2004.

There were a number of consequences for IRFs as a result of the implementation of TEFRA and the acute hospital PPS. First, earlier discharges from acute care hospitals led to greater demand for PAC facilities.8 As acute care hospitals began to discharge patients “quicker and sicker”—a function of the PPS incentive to control costs by reducing LOSs9—IRFs acted as an intermediate step between discharge from acute care hospitals and a permanent living situation.8, 10 Hospitals opened IRF units to retain the patients they were discharging, and new freestanding IRF hospitals were opened.11

Second, TEFRA exerted pressure on IRFs to control LOS. IRFs received a bonus payment when their costs for providing care were less than the “TEFRA limit,” a facility-specific cap on the average amount per discharge. When IRFs' costs were below their TEFRA limit, they received a payment equal to 50% of the difference between their costs and the cap, up to a maximum of 5% of the TEFRA limit. There was clearly an incentive to reduce LOS.12, 13 In addition, because these limits were calculated over a year of operation, facilities could offset the high costs incurred by patients with long LOS against lower LOS (and costs) for others. Averaging over an entire year allowed facilities to manage their costs through case-mix adjustments. Despite declining LOS, Medicare payments to IRFs grew, mainly because of increases in patient volume12 and increases in the IRF beds available through growth in the number of new IRFs.11

Third, IRFs that opened after the passage of TEFRA had a clear financial advantage.14, 15 IRFs existing at the onset of TEFRA had their base payments established in an earlier year. Once an IRF reached its cap, it was constrained to the annual Medicare payment updates. Newly opened IRFs, however, were exempt from base year payments for the first 3 years of operation. Their base payments were determined in the third year of operation, allowing them to open, and continue to operate, with higher costs than those faced by older IRFs. Thus, there was clearly an incentive to open new IRFs under TEFRA.11, 13 By the mid-1990s, PAC was the fastest growing sector of the Medicare program.8, 16, 17

The 1997 BBA, which introduced prospective payment to IRFs, was implemented to control the rapid expansion of Medicare expenditures in PAC.18, 19 After BBA, newly opened IRFs were capped in the amount they could be reimbursed, based on prevailing payments, thus lowering the incentive to open new, higher-cost IRFs. With the implementation of the PPS in 2002, IRFs received a predetermined payment per discharge based on a patient's functional level, as well as market wage adjustments that were unrelated to their own costs, with the exception of adjustments for cost outliers, and for a disproportionate share of low-income patients. Thus, the incentive to reduce LOS that existed under TEFRA intensified under PPS.19, 20 IRF costs per discharge are highly correlated to LOS, so managing LOS is a key to containing costs and maximizing reimbursement.5, 21 This incentive is especially important under PPS, because 24-hour access to physician and nursing care, and the 3-hour-a-day therapy minimum that patients must receive, limit the ability of IRFs to reduce daily costs. Consequently, IRF LOSs have declined since the implementation of PPS,22 accompanied by commensurate pressure to increase the number of patient admissions in order to maintain traditionally high occupancy rates.

Despite the significant changes in how IRFs are reimbursed by Medicare, one of their major payers, little is known about the impact of these changes on IRF supply and use. This study focuses on the availability of IRFs as reflected in the openings and closures of IRF facilities, the number of IRF beds, and inpatient days used between 1996 and 2004. We posed 4 questions: (1) What were the overall trends in IRF openings and closures during this period? (2) What were the overall trends in number of IRF beds and total inpatient days? (3) Did the trends vary depending on IRF characteristics? (4) Did the onset of IRF PPS have an effect on openings, closures, number of beds, and inpatient days, after controlling for the overall trend?

IRF PPS, like TEFRA before it, creates incentives for IRFs to reduce LOS. This creates pressure for increased admissions in order to maintain occupancy rates. However, because providers must manage individual patients' LOS, it creates a stronger incentive than under TEFRA. Pressure for increased admissions creates competition for patients among providers. McCue and Thompson5 found early adopters of IRF PPS did not increase admissions, probably because early in 2002, IRFs were operating at full capacity. However, continued declines in LOS23 and increasing competition from other PAC providers increased pressure to maintain census. When a facility cannot sustain sufficient admissions or reduce operating costs sufficiently to offset a reduction in revenues from lower admissions, its financial viability is at risk. In addition, profit status affects a facility's response to risk in that for-profit IRF owners, unlike not-for-profit board members, are directly responsible for losses and directly benefit from sales of property and facilities after closures. Thus, for-profit owners experience a direct financial incentive to close or open IRFs in response to changes in net profits expected under PPS. IRFs with fewer beds are expected to have lower costs related to either opening or closing and therefore to be more flexible in responding to changing financial conditions.

Given these considerations, we tested 2 hypotheses regarding overall trends and the effects of PPS on these trends: (1) there was a significant downward trend over the 1996 to 2004 period in the probability of openings, beds, and inpatient days and an upward trend in the probability of closures; and (2) the impact of IRF PPS is to intensify the likelihood of these trends.

Data Sources 

The HCRIS Hospital Cost Reports from 1996 through 2004 serve as the primary data for these analyses. Cost reports provide a summary of Medicare-covered hospital activities. Although cost reports may be reopened or audited, Cowles24 found that audits and reviews made almost no impact on variables included in this study such as inpatient days. For the purposes of this study, hospitals were considered open (operating) during a given year if a cost report was submitted with a starting date between October 1 of the previous year and September 30 of that year. Similarly, the IRF-specific data are based on the 12 months covered by the cost reports. Most hospital fiscal years start July 1, and thus the cost report data usually cover half of the current calendar year and half of the following calendar year. For example, what we report as data for the year 2004 for a given hospital with a fiscal year start date of July 1, 2004, would include data for half of calendar year 2004 and half of calendar year 2005.

We confirmed the operating status of hospitals with missing cost reports through internet searches and telephone inquiries. Missing cost reports were a somewhat larger problem in 2004 (191 missing) possibly because of lags in when cost reports are available in the database. Therefore, we used 2004 Medicare claims data to identify IRFs with Medicare claims that did not have cost reports in that year. We verified the status of these IRFs during 2004 using web searches and telephone calls. Using these strategies, we accounted for all but 1 missing provider ID in 2004. A facility may not operate as an IRF in a given year because it was out of compliance. When this information was available, we coded the hospital as closed. Finally, in 4.4% of observations, either bed size or patient days were missing for a given year. When these data were unobtainable by other strategies, values were assigned based on best available data for that measure in the closest year. Given the very small percentage of cases for which data were unavailable, we do not expect these to affect overall findings significantly.

The cost report data also provide information on location (ie, zip code) and control status of the hospital (for-profit or not-for-profit, including government hospitals). The organizational status (freestanding or unit) is obtained from the Medicare provider identification number. Hospitals with a 30 in the third and fourth positions of the 6-character identifier are freestanding rehabilitation hospitals. A rehabilitation unit takes the same provider number as the parent hospital but has the letter T in the third position. Maryland is a waiver state; its hospitals are not paid under PPS. Consequently, hospitals in this state are not included in the analyses.

Area resource files were used to provide countywide data on the number of SNFs in the area, population density (population a square mile), percent of population over age 65 years, and percent of the Medicare population in a Medicare HMO. We merged each IRF's data based on the county in which the IRF was located. These data were used to control further for the market characteristics in which an IRF operated.

Limits of Data on IRF Openings and Closures 

We identified Medicare-certified hospitals by the provider ID number. We defined openings as the appearance of new provider numbers in a given year relative to the previous year. We defined closures as the absence of a provider number relative to the previous year. Because the focus of our study was whether a hospital or unit was operating as an IRF in a given year, we did not assume when a provider ID was absent that had been present in the previous year that the facility was not operating. When hospitals merge, one facility may assume another facility's provider ID, and it appears as though 1 facility is no longer operating. In these cases, we confirmed the hospital's operating status. When hospitals merged and both hospitals continued to operate, we report both hospitals as open. There were some complex situations such as a merger in which hospital A assumed hospital B's provider ID but in fact hospital B closed its rehabilitation unit, not hospital A. To the extent possible, we documented such cases. While we may have missed some events, we were able to account for 99.9% of provider IDs. Thus, we expect the number of missed opening and closures is small. We provide specific cases of actual changes of status and how we coded these events.

When 2 hospitals merge, they are included in the HCRIS database under a single Medicare provider number and address. This Medicare provider ID may be one of the previous hospital's provider numbers, or it may be a new one. Thus, IRFs may change their provider ID even though they are operating as they were prior to the merger. For example, from 1996 to 1999, Faxton-St. Luke's Healthcare system in Utica, NY, operated an IRF subprovider at Faxton Hospital. In 2000, Faxton Hospital and nearby St. Luke's Memorial Hospital, formerly part of a cooperative affiliation, consolidated into 1 healthcare system and took on the St. Luke's Memorial provider ID. Examining changes in provider IDs only, it would appear as though Faxton Hospital and its IRF unit closed in 2000, and St. Luke's Memorial opened its own IRF unit. However, we confirmed that the IRF unit at Faxton Hospital remained open after this merger, and that no IRF unit exists at the St. Luke's Memorial campus. In this case, we coded Faxton Hospital's unit as open each year, and did not code a unit opening under St. Luke's Memorial ID.

In some cases, 2 facilities merged and took on 1 provider ID, but both continued to operate separate IRF units. In this situation, we report each IRF unit as open in that year. However, some mergers resulted in the closure of 1 of the merging hospitals' IRF units. In 1999, for example, the Orlando Regional Medical Center in Orlando, FL, purchased nearby Lucerne Medical Center. Both facilities operated IRF units at the time of the merger, but the Orlando Regional Medical Center unit subsequently closed, transferring its 18 beds to the Lucerne campus. While it appears from the provider ID that Orlando Regional Medical Center still operated an IRF unit, the Lucerne unit appeared to close. We coded Orlando Regional Medical Center as closed in 2000 and Lucerne's IRF unit as open for that year.

It is possible that a freestanding IRF, located on the same campus as an acute care hospital, will become a subprovider of that hospital although there were no physical changes in their status. In cases of a change from freestanding to unit status, we coded this as a closure of the freestanding hospital and an opening of a new unit. Reporting the change may or may not signify an important change in organizational control and financial incentives. Freestanding IRFs may also convert to LTCHs or SNFs. We code this as a closure of the IRF (converted).

Medicare requires that facilities seeking initial certification as an LTCH facility must demonstrate an average LOS of at least 25 days over a minimum of 6 months. Thus, many LTCHs open as Medicare-certified IRFs and convert to an LTCH within a year. We do not include these IRFs in our IRF data set because their dramatically longer than average LOS suggest the patients they admit and the organization of services are quite different from typical IRFs. We report separately the number of IRFs that opened and converted in this way.

Our approach differs from that of others23, 25 who report the number of open facilities based only on Medicare provider IDs. Thus, for the reasons discussed, the number of IRFs we report in any given year may differ slightly from those reported elsewhere. Reporting facilities as open after merging under a single Medicare provider ID tends to raise our reported numbers, while not including facilities that converted to LTCH status within 12 months of opening tends to lower our reported numbers compared with published reports. We believe that our approach is consistent with our objective of describing changes in the IRF industry from 1996 to 2004.

Data Analysis 

We provide cross-sectional and trend data on 4 outcomes: (1) the number of IRF openings, (2) the number of IRF closures, (3) changes in the number of IRF beds, and (4) the number of IRF inpatient days over the period. Because of potentially different service goals and costs across different types of IRFs, we distinguish between freestanding or unit, facilities, profit or not-for-profit IRFs, and sizes of the IRFs. Because the distribution of IRFs and regulations regarding IRFs vary geographically, we examined changes in the outcomes by census region and urban or rural settings. Because local market forces may influence these outcomes, we included county-level factors such as population per square mile, percent of the population over age 65 years, percent of Medicare population in Medicare HMOs, and number of SNFs in the county.

We compared trends from 1996 to 2002 and from 2002 to 2004 to examine changes between pre- and post-PPS implementation periods. Because we wish to examine the effects of PPS beyond overall industry trends, we used regression analysis to estimate both the statistical significance of trends from 1996 to 2004 and additional effects of PPS. We do this by including a trend variable (that takes the value of −6 in 1996 and increases 1 unit each year); a PPS variable that takes the value of 1 if the observation takes place after the introduction of a PPS, for 2002, 2003, or 2004; and a trend-PPS interaction term. We constructed these variables so that the coefficient of the trend variable measures the annual change in the outcome variable over the period prior to the PPS (1996–2001), the PPS coefficient measures the additional increase (or decrease) in the outcome in 2002 relative to what was expected from the overall trend, and the trend-PPS interaction coefficient measures the change in the slope of the trend line from 2002 to 2003 and 2003 to 2004.

To explain the probability of an IRF closing and of an IRF being a new IRF in any year, we estimated random-effects logistic regression models. For closures, we estimated how an IRF's characteristics in any year affect its probability of closing in the next year. For openings, we estimate how an IRF's characteristics in any year affect its probability of being a new IRF (having opened in that year). IRFs are no longer in the sample after they close or before they open, respectively. Thus, an IRF open in all years would contribute 8 observations, 1 for each year, beginning in 2004. An IRF that closed in 1998 would contribute 1 observation in the odds of closure regression model. For the openings model, IRFs that opened in 2004 would contribute only 1 observation. IRFs that opened after 1996 were not included in the models as potential closures, and IRFs that closed prior to 2005 were not included as potential new IRFs.

To estimate the probability of an IRF opening, we estimated the probability of a given IRF in a given year being a new IRF rather than an IRF that was already opened in the prior year—for example, the probability of an IRF opening means the probability of an IRF being a new IRF. The random-effects specification allows for a correlation between observations for a given IRF across years. Because of the complex specification of trend-PPS interactions and the nonlinear relationship of relative odds reported to the actual probabilities of opening and closing, we provide figures showing the regression model's predicted probabilities of openings and closures in each year given the mean values of IRF characteristics.

In addition, for IRFs that operated continuously, 1996 to 2004, we estimated fixed-effects ordinary least-squares regression models to explain number of beds and total inpatient days. The fixed-effect model controls for average differences observed between IRFs and controls for fixed characteristics of IRFs such as control status, geographic location, and whether they are freestanding or units. Geographic variables that are fixed over the entire period are eliminated from the number of beds and inpatient days regression models, population density is omitted because of its limited variation over the period, and bed size is eliminated because it is one of the dependent variables.

Openings and Closures 

In 1996, there was a total of 1037 IRFs, of which 185 were freestanding (17.8%) and 852 were units (82.2%). By 2004, there was a total of 1235 IRFs, including 214 freestanding IRFs (17.3%) and 1021 IRF units (82.7%). Table 1 illustrates the year-to-year changes in the number of freestanding IRFs and IRF units because of openings and closures. Not shown are 30 freestanding IRFs that opened and subsequently converted to an LTCH within 12 months. We did not include these facilities in the tables or regression analyses. In addition to the 21 freestanding IRF closures, another 22 freestanding IRFs closed and converted to an LTCH SNF (after operating as a freestanding IRF for more than 12 months). A total of 146 IRF units closed between 1996 and 2004.

Table 1. Openings and Closures of IRFs for the Years 1996 Through 2004

	IRFs	Pre-PPS						Post-PPS
		1996	1997	1998	1999	2000	2001	2002	2003	2004
	Total operating⁎
	Freestanding	185(17.8)	192(17.5)	193(16.8)	190(16.4)	192(16.4)	200(16.9)	205(17.1)	217(17.7)	214(17.3)
	Unit	852(82.2)	908(82.5)	955(83.2)	972(83.6)	982(83.6)	983(83.1)	993(82.9)	1012(82.3)	1021(82.7)
	Total	1037	1100	1148	1162	1174	1183	1198	1229	1235
	Existing†
	Freestanding		182(17.8)	186(17.2)	185(16.4)	185(16.3)	187(16.3)	191(16.6)	203(17.3)	212(17.5)
	Unit		839(82.2)	894(82.8)	942(83.6)	948(83.7)	959(83.7)	957(83.4)	970(82.7)	1002(82.5)
	Total		1021	1080	1127	1133	1146	1148	1173	1214
	Openings‡
	Freestanding		10(12.7)	7(10.3)	5(14.3)	7(17.1)	13(35.1)	14(28.0)	14(25.0)	2(9.5)
	Unit		69(87.3)	61(89.7)	30(85.7)	34(82.9)	24(64.9)	36(72.0)	42(75.0)	19(90.5)
	Total		79	68	35	41	37	50	56	21
	Closure§
	Freestanding		1(6.3)	3(15.0)	5(23.8)	2(6.9)	3(10.7)	4(11.4)	1(4.0)	2(13.3)
	Converted∥		2(12.5)	3(15.0)	3(14.3)	3(10.3)	2(7.1)	5(14.3)	1(4.0)	3(20.0)
	Unit		13(81.3)	14(70.0)	13(61.9)	24(82.8)	23(82.1)	26(74.3)	23(92.0)	10(66.7)
	Total		16	20	21	29	28	35	25	15

NOTE. Values are openings (closures).

⁎ All IRFs open in this fiscal year, including existing hospitals and new hospitals opened that year. † Hospital or unit was operating in the previous fiscal year. ‡ Hospital or unit began operating in this fiscal year. § Hospital or unit ceased operating in this fiscal year. ∥ Facilities that converted from IRF to another type of facility.

Table 2 describes the number of freestanding and unit IRFs in 1996, 2002, and 2004 by profit status, size, location (urban or rural), and census region. There is a notable difference in the average size of freestanding IRFs and IRF units. The median size for freestanding IRFs is 57 beds in 1996 (IQR, 40–80; mean size, 64.8 beds), and for IRF units the median size is 20 beds (IQR, 15–30; mean size, 24.9 beds). Overall, there was a 15.7% growth in freestanding IRFs (185–214), somewhat less than the 19.8% growth in IRF units (852–1021) over the period. There was comparable percent growth in the for-profit sector of both IRF units and freestanding IRFs. In contrast, not-for-profit freestanding IRFs experienced a small net decline (−10.1%) in facilities, while not-for-profit IRF units saw a moderate growth.

Table 2. Comparison of Freestanding IRFs and IRF Units in 1996, 2002, and 2004 by Profit Status, Size, Location, and Census Region

	Categories	1996	2002	2004	% Change 1996–2004	Absolute Change (Annual % Change) 1996–2002	Absolute Change (Annual % Change) 1996–2002
	Operating IRFs, n (%)
	Freestanding	185(17.8)	205(17.1)	214(17.3)	29(15.7)	20(1.7)	9(2.2)
	Unit	852(82.2)	993(82.9)	1021(82.7)	169(19.8)	141(2.6)	28(1.4)
	Total	1037	1198	1235
	Profit status, ⁎n (%)
	For-profit
	Freestanding	116(43.6)	141(42.6)	152(42.8)	36(31.0)	25(3.3)	11(3.8)
	Unit	150(56.4)	190(57.4)	203(57.2)	53(35.3)	40(4.0)	13(3.4)
	Total	266	331	355
	Not-for-profit
	Freestanding	69(8.9)	63(7.3)	62(7.0)	−7(−10.1)	−6(−1.5)	−1(−0.8)
	Unit	702(91.1)	803(92.7)	818(93.0)	116(16.5)	101(2.3)	15(0.9)
	Total	771	866	880
	Size,† n (%)
	Small‡
	Freestanding	91(20.5)	83(16.0)	90(16.5)	−1(−1.1)	−8(−1.5)	7(4.1)
	Unit	352(79.5)	437(84.0)	455(83.5)	103(29.3)	85(3.7)	18(2.0)
	Total	443	520	545
	Large‡
	Freestanding	94(15.8)	122(18.0)	124(18.0)	30(31.9)	28(4.4)	2(0.8)
	Unit	500(84.2)	556(82.0)	566(82.0)	66(13.2)	56(1.8)	10(0.9)
	Total	594	678	690
	Location,§ n (%)
	Urban
	Freestanding	171(18.8)	188(18.6)	193(18.8)	22(12.9)	17(1.6)	5(1.3)
	Unit	737(81.2)	825(81.4)	836(81.2)	99(13.4)	88(1.9)	11(0.7)
	Total	908	1013	1029
	Rural
	Freestanding	14(10.9)	17(9.2)	21(10.2)	7(50.0)	3(3.3)	4(11.1)
	Unit	115(89.1)	168(90.8)	185(89.8)	70(60.9)	53(6.5)	17(4.9)
	Total	129	185	206
	Census region,∥ n (%)
	Northeast
	Freestanding	39(22.0)	44(20.0)	45(19.9)	6(15.4)	5(2.0)	1(1.1)
	Unit	138(78.0)	176(80.0)	181(80.1)	43(31.2)	38(4.1)	5(1.4)
	Total	177	220	226
	Midwest
	Freestanding	22(7.3)	26(8.0)	27(8.0)	5(22.7)	4(2.8)	1(1.9)
	Unit	279(92.7)	299(92.0)	309(92.0)	30(10.8)	20(1.2)	10(1.7)
	Total	301	325	336
	South
	Freestanding	92(24.7)	109(24.0)	116(24.6)	24(26.1)	17(2.9)	7(3.2)
	Unit	280(75.3)	346(76.0)	356(75.4)	76(27.1)	66(3.6)	10(1.4)
	Total	372	455	472
	West
	Freestanding	32(17.1)	26(13.1)	25(12.5)	−7(−21.9)	−6(−3.4)	−1(−1.9)
	Unit	155(82.9)	172(86.9)	175(87.5)	20(12.9)	17(1.7)	3(0.9)
	Total	187	198	200

⁎ Worksheet S2, line 18.33 † Worksheet S3, column 1.33 ‡ Small: Freestanding IRFs=fewer than 57 beds, IRF units=fewer than 20 beds; large: freestanding IRFs=57 or more beds, IRF units=20 or more beds. § Worksheet S2, line 21.33 ∥ Hospital state from Worksheet S2, line 1.01, applied to US Census regions.33

Over the period, the largest growth occurred for large freestanding IRFs (30.3%); for-profit IRFs, whether freestanding or units; and rural IRFs, whether freestanding (50%) or units (60.9%). However, the large growth of rural freestanding IRFs represented only 7 new openings. By region, the northeast saw the most growth in IRF units (31.2%) but less growth in freestanding IRFs. The midwest and south saw modest percent growth in both IRF units and freestanding IRFs, although the actual number of new facilities was significantly greater in the south. By contrast, the west experienced a loss of freestanding IRFs (almost 22%) and only modest growth in IRF units (12.9%).

Figure 1 shows the distribution of IRFs in the 48 contiguous states, coded by whether they were continuously open or they opened, closed, or converted during the period. This figure shows growth in rural areas and other traditionally underserved areas. The map highlights a concentration of conversions in the south and midwest.

View Large Image Download to PowerPoint

Fig 1. Status of IRF freestanding hospitals and units, 1997 to 2004.

Number of IRF Beds and Patient Days 

Openings and closures provide 1 indicator of organizational change and also of changes in the supply of IRF services. However, because of the large variation in IRF size, it is important to examine other indicators of access and use of IRF services. The number of IRF beds represents all beds available for use by patients at the end of the cost-reporting period.26 It is an indicator of the short-term capacity of IRFs in a given year. Total bed days available is the sum of the number of days each bed is available throughout the cost-reporting period, multiplied by the number of beds. It is an estimate of the capacity of an IRF in a given year. Total inpatient days represent the actual count of days that patients used hospital beds. It is the number of discharges multiplied by length of stay. Occupancy rates represent the extent to which facilities used the available bed days. They are calculated by dividing total inpatient days by total bed days.

We do not report the number of discharges each year, because interrupted stays were reported differently before and after the introduction of IRF PPS. Interrupted stays occur when a patient is discharged and returns to the same IRF within 3 days. Prior to PPS, this was reported to Medicare as 2 rehabilitation admissions. After PPS, this was considered a single admission, making it appear that the number of discharges decreased in the PPS period relative to the pre-PPS period even if there was no real change.

In contrast with a count of the number of IRFs operating, the number of beds and total inpatient days provide an indication of IRF capacity over the entire cost-reporting period and the extent to which capacity was used. This is important, given that many openings and closures were concentrated among small freestanding facilities in a few states. Louisiana had 42 openings and 15 closures over the period, representing 10.9% of all openings and 7.6% of all closures in the period. Louisiana's changes included 25 openings and 9 closures of freestanding IRFs. However, the average size of the freestanding IRFs that opened or closed was only 25 beds and 30 beds, respectively.

Table 3 shows the number of IRF beds in 1996 and 2004. Overall, there were 33,166 beds in 1996, of which 63.8% were in IRF units. While only 17.8% of IRF facilities were freestanding in 1996, 36.2% of IRF beds were in freestanding IRFs in that year, reflecting the much larger bed size of freestanding IRFs. The total number of beds grew to 38,386 by 2004, an overall growth of 15.7%. Percent growth was about the same for IRF units and freestanding IRFs at 16.5% and 14.4%, respectively. Freestanding IRF beds dominate the for-profit market, while IRF unit beds dominate the not-for-profit sector, reflecting the predominance of nonprofits among acute care hospitals. This effect increased over the period, with a 15.2% growth in not-for-profit IRF unit beds and a 7.5% loss of not-for-profit IRF freestanding beds. The for-profit market grew moderately over the period for both freestanding IRFs beds and unit IRF beds. Small freestanding IRFs (fewer than 57 beds) saw a modest loss of IRF beds (−15.7%), while larger freestanding IRFs and large and small units saw moderate growth in bed size. IRF beds are heavily concentrated in urban/suburban areas (92.2%). Rural IRF units had considerable growth (48.3%). By contrast, rural freestanding IRFs lost beds pre-PPS, but had growth in beds thereafter for an overall gain of 13.4%.

Table 3. Total Number of Beds in 1996, 2002, and 2004

	Categories	1996	2002	2004	% Change 1996–2004	Annual % Change 1996–2002	Annual % Change 2002–2004
	Operating IRFs, n (%)
	Freestanding	11,996(36.2)	13,281(0.4)	13,729(35.8)	1733(14.4)	1285(1.7)	448(1.7)
	Unit	21,170(63.8)	24,161(0.6)	24,657(64.2)	3487(16.5)	2991(2.2)	496(1.0)
	Total	33,166	37,442	38,386
	Profit status,⁎ n (%)
	For-profit
	Freestanding	7133(68.1)	8761(0.7)	9233(69.1)	2100(29.4)	1628(3.5)	472(2.7)
	Unit	3346(31.9)	4089(0.3)	4120(30.9)	774(23.1)	743(3.4)	31(0.4)
	Total	10,479	12,850	13,353
	Not-for-profit
	Freestanding	4863(21.4)	4503(0.2)	4496(18.0)	−367(−7.5)	−360(−1.3)	−7(−0.1)
	Unit	17,824(78.6)	20,072(0.8)	20,537(82.0)	2713(15.2)	2248(2.0)	465(1.2)
	Total	22,687	24,575	25,033
	Size,† n (%)
	Small‡
	Freestanding	3514(42.4)	2776(0.3)	2961(32.7)	−553(−15.7)	−738(−3.9)	185(3.3)
	Unit	4778(57.6)	5877(0.7)	6105(67.3)	1327(27.8)	1099(3.5)	228(1.9)
	Total	8292	8653	9066
	Large‡
	Freestanding	8482(34.1)	10,505(0.4)	10,768(36.7)	2286(27.0)	2023(3.6)	263(1.2)
	Unit	16,382(65.9)	18,284(0.6)	18,552(63.3)	2170(13.2)	1902(1.8)	268(0.7)
	Total	24,864	28,789	29,320
	Location,§ n (%)
	Urban
	Freestanding	11,458(37.5)	12,837(0.4)	13,117(37.7)	1659(14.5)	1379(1.9)	280(1.1)
	Unit	19,135(62.5)	21,426(0.6)	21,639(62.3)	2504(13.1)	2291(1.9)	213(0.5)
	Total	30,593	34,263	34,756
	Rural
	Freestanding	538(20.9)	444(0.1)	612(16.9)	74(13.8)	−94(−3.1)	168(17.4)
	Unit	2035(79.1)	2735(0.9)	3018(83.1)	983(48.3)	700(5.1)	283(5.0)
	Total	2573	3179	3630
	Census region,∥ n (%)
	Northeast
	Freestanding	3759(50.1)	4005(0.5)	3998(45.1)	239(6.4)	246(1.1)	−7(−0.1)
	Unit	3741(49.9)	4700(0.5)	4872(54.9)	1131(30.2)	959(3.9)	172(1.8)
	Total	7500	8705	8870
	Midwest
	Freestanding	1404(17.5)	1776(0.2)	1804(20.8)	400(28.5)	372(4.0)	28(0.8)
	Unit	6634(82.5)	6873(0.8)	6863(79.2)	229(3.5)	239(0.6)	−10(−0.1)
	Total	8038	8649	8667
	South
	Freestanding	5299(41.3)	5995(0.4)	6458(41.6)	1159(21.9)	696(2.1)	463(3.8)
	Unit	7524(58.7)	8920(0.6)	9062(58.4)	1538(20.4)	1396(2.9)	142(0.8)
	Total	12,823	14,915	15,520
	West
	Freestanding	1534(31.9)	1505(0.3)	1469(27.6)	−65(−4.2)	−29(−0.3)	−36(−1.2)
	Unit	3271(68.1)	3668(0.7)	3860(72.4)	589(18.0)	397(1.9)	192(2.6)
	Total	4805	5173	5329

⁎ Worksheet S2, line 18.33 † Worksheet S3, column 1.33 ‡ Small: Freestanding IRFs=fewer than 57 beds, IRF units=fewer than 20 beds; large: freestanding IRFs=57 or more beds, IRF units=20 or more beds. § Worksheet S2, line 21.33 ∥ Hospital state from Worksheet S2, line 1.01, applied to US Census regions.33

Table 3 also shows IRF beds by region. In the northeast, there was fairly even distribution of IRF unit and freestanding IRF beds in 1996, but IRF units had much greater growth in number of beds over the period, 30.2%, compared with 6.4% for freestanding IRFs. In the midwest, IRF care was heavily dominated by beds in IRF units in 1996, although freestanding IRFs saw greater growth over the period, 28.5%, compared with 3.5% for IRF units. In the south, there were somewhat more IRF unit beds than freestanding IRF beds in 1996, and both saw approximately the same moderate growth over the period, so the distribution of beds remains essentially unchanged between IRF units and freestanding IRFs in 2004. In the West, there were twice as many IRF unit beds as freestanding IRF beds in 1996. This distribution was accentuated by 2004, with 4.2% fewer freestanding IRF beds and an 18% increase in IRF unit beds over the period. The annual percent change in beds was fairly consistent with a few notable exceptions. For-profit units showed no growth after 2002, while rural freestanding IRFs reversed a decline in beds after 2002.

Bed Growth as a Result of IRF Openings and Closures, and Growth of Existing IRFs 

Figure 2 shows the percent change in the number of beds between 1996 and 2004, disaggregated by gains due to newly opened IRFs during the period, losses because of closed IRFs, and change caused by increased or decreased bed size of IRFs present over the entire period. In each case, we reported the percentage growth relative to the total number of beds in 1996. Thus, the overall growth in beds of 15.7% over the period was the result of a loss of 12.5% of 1996 beds because of IRF closures, an increase of 23.1% because of IRF openings, and an increase of 5.1% in the number of beds because of the growth of existing IRFs. Openings were more responsible for overall bed growth than was growth of existing facilities. While similar results hold for both freestanding and units, freestanding had higher bed growth as a result of the growth of existing facilities and lower growth as a result of new facilities.

View Large Image Download to PowerPoint

Fig 2. Growth in IRF beds, 1996 to 2004.

Table 4 presents the total number of inpatient days. Total inpatient days increased from 7,999,363 days to 9,032,110 days over the period, a growth of 12.9%, somewhat less than the 15.7% growth in beds. There was little change in the percentage of total patient days across IRFs, with 62% and 63% of days spent in IRF units in 1996 and 2004, respectively. Most for-profit, inpatient days were in freestanding IRFs, while most not-for-profit inpatient days were in IRF units. For-profit and large freestanding IRFs had a decline in inpatient days, while increasing beds after the PPS. By contrast, not-for-profit freestanding IRFs reduced beds along with inpatient days after PPS. Rural freestanding hospital beds grew much faster than inpatient days. By region, all freestanding IRFs had a loss of inpatient days after PPS. The largest decline was in the northeast. Inpatient days in units generally held stable or had a little growth across regions.

Table 4. Description of Number of Total Inpatient Days, 1996, 2002, and 2004

		1996	2002	2004	% Change 1996–2004	Annual % Change 1996–2002	Annual % Change 2002–2004
	Operating IRFs, n (%)
	Freestanding	3,022,275(37.8)	3,533,620(38.4)	3,311,917(36.7)	289,642(9.6)	511,345(2.6)	−221,703(−3.2)
	Unit	4,977,088(62.2)	5,678,523(61.6)	5,720,193(63.3)	743,105(14.9)	701,435(2.2)	41,670(0.4)
	Total	7,999,363	9,212,143	9,032,110
	Profit Status,⁎ n (%)
	For-profit
	Freestanding	1,792,464(71.6)	2,376,122(73.2)	2,173,898(72.4)	381,434(21.3)	583,658(4.8)	−202,224(−4.3)
	Unit	711,048(28.4)	868,124(26.8)	827,808(27.6)	116,760(16.4)	157,076(3.4)	−40,316(−2.3)
	Total	2,503,512	3,244,246	3,001,706
	Not-for-profit
	Freestanding	1,229,811(22.4)	1,156,635(19.4)	1,138,019(18.9)	−91,792(−7.5)	−73,176(−1.0)	−18,616(−0.8)
	Unit	4,266,041(77.6)	4,810,400(80.6)	4,892,385(81.1)	626,344(14.7)	544,359(2.0)	81,985(0.8)
	Total	5,495,852	5,967,035	6,030,404
	Size,† n (%)
	Small‡
	Freestanding	830,336(44.0)	672,259(34.0)	706,493(34.6)	−123,843(−14.9)	−158,077(−3.5)	34,234(2.5)
	Unit	1,054,877(56.0)	1,305,198(66.0)	1,336,908(65.4)	282,031(26.7)	250,321(3.6)	31,710(1.2)
	Total	1,885,213	1,977,457	2,043,401
	Large‡
	Freestanding	2,191,939(35.9)	2,861,361(39.6)	2,605,424(37.3)	413,485(18.9)	669,422(4.5)	−255,937(−4.6)
	Unit	3,922,212(64.1)	4,373,324(60.4)	4,383,285(62.7)	461,073(11.8)	451,112(1.8)	9961(0.1)
	Total	6,114,151	7,234,685	6,988,709
	Location,§ n (%)
	Urban
	Freestanding	2,932,208(39.1)	3,447,629(40.3)	3,215,448(38.6)	283,240(9.7)	515,421(2.7)	−232,181(−3.4)
	Unit	4,559,300(60.9)	5,117,750(59.7)	5,110,991(61.4)	551,691(12.1)	558,450(1.9)	−6759(−0.1)
	Total	7,491,508	8,565,379	8,326,439
	Rural
	Freestanding	90,067(17.7)	85,991(13.3)	96,469(13.7)	6402(7.1)	−4076(−0.8)	10,478(5.9)
	Unit	417,789(82.3)	560,774(86.7)	609,202(86.3)	191,413(45.8)	142,985(5.0)	48,428(4.2)
	Total	507,856	646,765	705,671
	Census region,∥ n (%)
	Northeast
	Freestanding	1,030,127(50.6)	1,127,323(48.4)	969,039(44.7)	−61,088(−5.9)	97,196(1.5)	−158,284(−7.3)
	Unit	1,007,426(49.4)	1,201,393(51.6)	1,198,425(55.3)	190,999(19.0)	193,967(3.0)	−2968(−0.1)
	Total	2,037,553	2,328,716	2,167,464
	Midwest
	Freestanding	321,135(17.8)	426,264(22.0)	419,487(21.5)	98,352(30.6)	105,129(4.8)	−6777(−0.8)
	Unit	1,480,968(82.2)	1,514,777(78.0)	1,532,099(78.5)	51,131(3.5)	33,809(0.4)	17,322(0.6)
	Total	1,802,103	1,941,041	1,951,586
	South
	Freestanding	1,374,981(42.7)	1,604,677(42.7)	1,556,111(42.1)	181,130(13.2)	229,696(2.6)	−48,566(−1.5)
	Unit	1,842,394(57.3)	2,151,754(57.3)	2,136,013(57.9)	293,619(15.9)	309,360(2.6)	−15,741(−0.4)
	Total	3,217,375	3,756,431	3,692,124
	West
	Freestanding	296,032(31.4)	375,356(31.7)	367,280(30.1)	71,248(24.1)	79,324(4.0)	−8076(−1.1)
	Unit	646,300(68.6)	810,599(68.3)	853,656(69.9)	207,356(32.1)	164,299(3.8)	43,057(2.6)
	Total	942,332	1,185,955	1,220,936

⁎ Worksheet S2, line 18.33 † Worksheet S3, column 1.33 ‡ Small: Freestanding IRFs=fewer than 57 beds, IRF units=fewer than 20 beds; large: freestanding IRFs=57 or more beds, IRF units=20 or more beds. § Worksheet S2, line 21.33 ∥ Hospital state from Worksheet S2, line 1.01, applied to US Census regions.33

Because LOS was decreasing over this period,22, 27 facilities needed to increase the number of admissions if they were to offset the decreased LOS to maintain occupancy. Occupancy rates remained stable over the period. Freestanding IRFs consistently had a somewhat higher rate than units, ranging from a low of 70% in 1996 to a high of 75% in 2000. Under PPS, rates declined from 74% in 2002 to 71% in 2004. Occupancy rates in IRF units varied little over the period, from 67% in 1996, to 68% in 2002, to 66% in 2004.

Effect of PPS on Openings and Closures 

Table 5 presents the results of the logistic regressions, examining the effects of PPS on openings and closures of IRFs, after controlling for the linear trend in probability of openings and closures between 1997 and 2002. For closures, we estimated a random-effects logistic regression for 1035 IRFs and 7752 observations. For openings, the sample consisted of 1231 IRFs and 8797 observations. Thus, the closure regression model provides the predicted odds of closure, given an IRF was open in the previous year, while the openings regression provides the predicted odds that a hospital in a given year was a new rather than a pre-existing hospital in that year, given its characteristics and the year observed.

Table 5. Regression Analyses Impact of PPS on IRF Openings and Closures (1996–2004)

	Covariates	Odds Ratios
		Closures	Openings
	Trend	1.11	0.78‡
	PPS	1.30	2.34‡
	Trend⁎ PPS†	0.55‡	0.85
	For-profit	1.85‡	1.47‡
	Urban	0.84	0.62‡
	Freestanding	2.09⁎	2.86‡
	Bed size	0.98‡	0.96‡
	South	1.05	0.87
	West	0.83	0.55‡
	Midwest	1.27	0.69
	% Medicare HMO	1.02⁎	1.00
	% Population over age 65y	0.97	1.01
	Number of SNFs	1.00	1.00
	Population density	1.00	1.00
	Constant	0.05‡	0.08

⁎ Significant at P≤.01. † Interaction term between trend and PPS covariates. ‡ Significant at P≤.001.

The regression models control for-profit status, whether the IRF is freestanding or a unit, total number of IRF beds, whether it is in an urban area, region of the country, and county characteristics: population per square mile, percent of population over age 65 years, percent of Medicare population in HMOs, and number of SNFs in the county. Fig 3, Fig 4 present the regression results for closures and openings, respectively, for the mean values of each of the covariates—for example, controlling for the mean proportion of IRFs with each characteristic in 1996. Each figure also shows the unadjusted probability of closing and opening in each year.

View Large Image Download to PowerPoint

Fig 3. Probability of IRF closures, 1996 to 2004.

View Large Image Download to PowerPoint

Fig 4. Probability of IRF openings, 1996 to 2004.

Neither the trend over time nor the 1-time shift after the introduction of PPS was statistically significant in the closure model. However, after 2002, there was a statistically significant annual decline in the probability of closures. For openings, there was a statistically significant negative trend effect prior to PPS, with the predicted probability of an opening reduced by approximately 60% between 1997 and 2001. After 2001, there was a positive and statistically significant increase in the probability of openings with the introduction of the PPS, but the negative trend in probability of openings was maintained after 2002.

Figure 3 shows that the regression model does a reasonably good job of predicting the probabilities of IRF closures, with predicted and actual values showing the same year (2002) for the downturn in the probability of closures. Figure 4 shows that the actual probability of opening an IRF declined sharply from 1997 to 1999, increased between 1999 and 2003, and then declined sharply after 2003. This effect contrasts with our predicted regression model, in which we constrained the trends to be constant from 1996 to 2002 and from 2002 to 2004, but allowed for both a shift and change in trend (slope) starting in 2002. The predicted graph shows a gradual decline until 2001 with an increase in openings in 2002, followed by a decline. In contrast, an examination of the actual data shows it was the period just after the introduction of the BBA (1997–1999) and the last year of this study that most adversely affected the probability of new openings.

The IRF characteristics that significantly affected probability of closures over the entire period were for-profit status, freestanding IRFs, smaller IRFs, and IRFs in counties with more Medicare HMO penetration. The same factors explained the increased probability of openings, with the exception that rural location, rather than HMO penetration, was a significant factor.

Effect of PPS on Number of Beds and Total Inpatient Days 

For the 889 IRFs that operated every year between 1996 and 2004 (7993 observations), we estimated the effect of PPS on the number of beds and total inpatient days, controlling for the linear trend prior to PPS. Rather than control for location or other IRF characteristics, we estimated a fixed-effects linear regression, measuring only within IRF deviations from the average number of beds and total inpatient days for that IRF. This model controls for between-IRF differences that are described by the mean differences across IRFs (those differences that do not change over time).

For both beds and inpatient days, there was a statistically significant positive trend over the period, with the predicted number of beds per IRF increasing by .28 beds per year and the number of patient days per IRF increasing by 64 patient days per year between 1996 and 2002 (table 6). There was no statistically significant change in trend in number of beds after 2002, but total inpatient days had an overall decrease of 197 inpatient days per year (261 inpatient days per year less than the trend before PPS) after 2002. As shown in figure 5, the predicted values from the regression models show that the trend in bed growth remained positive after 2002, while the trend in total inpatient days, which increased prior to 2002, declined subsequently.

Table 6. Regressions Explaining Number of IRF Beds and Inpatient Days (1996–2004) for IRFs Operating Continuously Over the Entire Period

	Covariates	Coefficients
		Number of Beds	Total Inpatient Days
	Trend	.28⁎	64.26⁎
	PPS	−.43	−34.64
	Trend⁎ PPS†	−.10	−261.07⁎
	For-profit	.66	−42.26⁎
	Percent Medicare HMO	−.02	−25.52⁎
	Percent population over 65y	−.46⁎	−195.18⁎
	Number of SNFs	.00	−7.31
	Constant	40.61⁎	11737.38⁎

⁎ Significant at P≤0001. † Interaction term between trend and PPS covariates.

View Large Image Download to PowerPoint

Fig 5. Predicted number of IRF beds and total IRF inpatient days, 1996–2004.

Alternative Explanations 

The years of study represented a period of change for inpatient rehabilitation. While our focus was on the implementation of PPS, there were other factors influencing IRF behavior during that time. One concern is whether HealthSouth, a major provider of freestanding IRF care, had a disproportionate effect on openings and closures, representing approximately 40% of freestanding IRFs and 43% of freestanding IRF beds, by our estimates. HealthSouth openings represented about 30% of new freestanding IRFs in 1997 and 1998, about 15% of freestanding IRF openings in 2000 and 2001, and 36% in 2002. In 2003, 29% of freestanding IRF openings were by HealthSouth. Despite the much-publicized corporate difficulties that became public in 2003, our data do not suggest that this situation impacted the rate of new openings by HealthSouth in that year. In 2004, the industry overall ceased to open new IRFs30; there was only 1 new freestanding IRF opened in the US that year, and that was by HealthSouth.

CONs are programs by which states control the growth of hospitals. CON programs were federally mandated in 1974 and repealed in 1987. While a number of states eliminated their CON programs at that time, 26 states continue to regulate the construction of IRFs and IRF beds through CON programs.31 States with CON programs are distributed across the nation. Because we analyzed data by broad census regions, it is unlikely that CONs influenced our findings. Arizona, California, and Florida, states with high percentages of Medicare beneficiaries, do not have CON programs.

There is increased interest in MIS or minimal incision arthroplasty in which total joint replacements are conducted on an outpatient basis. However, the percentage of joint replacement surgeries conducted using MIS is still small and is unlikely to have played a role in reducing the number of inpatient days in the years covered by this study.

LTCHs provide rehabilitation for medically complex and fragile patients who require significantly longer stays than are typical in an IRF. Some IRFs may perceive LTCHs in the same market as competition, and in the past a few freestanding LTCHs have operated like IRFs. However, MedPAC30 has found that in areas with LTCHs, SNFs were the primary alternative to LTCH care.

Because IRFs were able to increase total inpatient days despite declining LOS by increasing admissions, Medicare payments to IRFs were more likely to have increased under a policy of payments tied to discharges.25 To the extent that greater admissions reflect increased patient access to rehabilitation care, this may be viewed as a positive result. Recently, there has been increased review of medical necessity by fiscal intermediaries and the 2003 Medicare Modernization Act implemented recovery audit contractors charged with recovery of overpayments. Both are strategies Medicare uses to monitor growing IRF admissions and payments.

Policy Implications 

These findings have important implications with regard to access and availability of IRF services in the United States. IRFs are not uniformly available in all markets, yet it is not clear that this is a reflection of demand for services that is associated with an aging population. The IRF PPS did little to affect service distribution in less well-served areas, although we did find growth in rural areas. In the western region, we saw a transition from freestanding IRFs to units. To the extent that changes in payment limit the availability of IRFs across the country, there are consequences for patients needing intensive rehabilitation services. PPS represents a major change in the way medical rehabilitation is funded, shifting much of the financial risk for patient care to the provider and creating a risk for decreased availability. Total inpatient days did not decline until the final year of observation. We postulate that this is related to more rigorous enforcement of the 75% rule than to a late effect of PPS. Occupancy rates have remained consistently high over the entire period (around 75%). Occupancy rates in 2004 were close to rates at the start of the period (70%). This observation implies that IRFs were implementing strategies to admit a sufficient number of patients, even though bed numbers were increasing and LOS was declining. Consequently, a policy that limits the potential of IRFs to increase patient admissions, such as the limits on admissions to IRFs of patients with conditions other than those included in the 75% rule,32 is likely to produce substantial decreases in total inpatient days.

Study Limitations 

Changes in organizations are not easy to categorize or model. Mergers, in which IRFs remain open in multiple places but they are reported to Medicare as a single entity, are an example. A freestanding IRF that becomes a unit of a hospital with no apparent change in its operations is another example. The organizational significance of these changes is unclear. However, we have attempted to be consistent in dealing with such changes and verifying when changes occurred.

It would have been preferable to report discharges and LOS in comparing the pre-PPS and PPS periods. However, changes in how interrupted stays were reported after PPS made this impossible with the study dataset. Given that others have documented that LOS was declining over this period,11, 22 maintaining or increasing total inpatient days can only be achieved by compensatory increases in admissions. Therefore, the inability to report changes in LOS is not critical to understanding the findings presented here.

Occupancy rates are strongly related to hospital profitability.29 There will be continued pressure on IRFs to maintain occupancy rates with patients from diagnostic groups covered under the 75% rule. Thus, we expect that greater organizational change will be observed in the period after 2004. Preliminary data suggest that it might be challenging for IRFs to offset fully the reduction in admissions that result from the 75% rule.