Trends in Length of Stay, Living Setting, Functional Outcome, and Mortality Following Medical Rehabilitation FREE

Length of stay (LOS) for inpatient medical rehabilitation has decreased dramatically in recent years^{1 - 6} and is expected to continue declining under the prospective payment system implemented for inpatient medical rehabilitation in January 2002.^{1 ,5} The impact that reduced LOS has had on rehabilitation outcomes including functional status, living setting, and mortality is unknown.

We examined changes in LOS for persons receiving inpatient medical rehabilitation from 1994 through 2001 using a large national database representative of rehabilitation patients in the United States.^{7 - 8} In addition to exploring trends in LOS for 5 major impairment groups (stroke, brain dysfunction, other neurologic conditions, spinal cord dysfunction [traumatic and nontraumatic], and orthopedic conditions), we also examined changes in rehabilitation effectiveness, efficiency, discharge to home, living setting at 3- to 6-month follow-up, and mortality. One goal was to establish baseline data for rehabilitation outcomes prior to the introduction of inpatient rehabilitation prospective payment by the Centers for Medicare & Medicaid Services (CMS) in 2002.^{4 - 6} We hypothesized that decreasing LOS would be associated with reduced functional status and decreased community living at follow-up.

The data were collected by the Uniform Data System for Medical Rehabilitation (UDSMR). The UDSMR is the largest national registry of standardized information on medical rehabilitation inpatients in the United States and has been used by rehabilitation facilities since 1987.^{7 - 8} The UDSMR database includes sociodemographic variables, diagnoses (International Classification of Diseases, Ninth Revision[ICD-9] codes), facility characteristics, patient prehospital living arrangements and marital status, predisability employment status, discharge disposition, and cost factors including LOS, source of payment, and hospital charges.

In addition to descriptive information for patients and facilities, scores on a standardized measure of basic daily living skills—the Functional Independence Measure (FIM Instrument)—are recorded at admission, discharge, and follow-up. The FIM instrument measures functional status using 18 items covering 6 domains: self-care or activities of daily living (6 items on dressing upper and lower body, eating, grooming, toileting, and bathing), bladder and bowel control (2 items), mobility (3 transfer items), locomotion (2 items on walking/wheelchair use and stairs), communication (2 items on comprehension and expression), and social cognition (3 items on social interaction, problem solving, and memory). All 18 items are scored into 1 of 7 levels of function ranging from complete dependence (level 1) to complete independence (level 7). Total scores range from 18 to 126, with higher scores indicating better function. Reliability, assessed using intraclass correlation coefficients, has consistently been found to be greater than 0.85.^{9 - 11}

Facilities follow the UDSMR protocol in administering the FIM instrument and subsequent submission of the data. The FIM instrument is administered within 72 hours after inpatient rehabilitation admission and 72 hours or less before discharge. Follow-up data obtained during the period of the study (1994-2001) were collected by the National Follow-up Services between 80 and 180 days after discharge using telephone interviews. Nurses trained in administering and interpreting the FIM instrument collected follow-up information. If the patient was not able to respond appropriately, the nurses used a protocol for collecting information from proxies (family member or caregiver).¹² In addition to the FIM items, information on living setting, employment status, outpatient therapy, marital status, and rehospitalization was collected at follow-up. The interview process has been described.^{12 - 13}

Follow-up information is aggregated with the complete UDSMR patient record. The interrater and test-retest reliability of the data collection process, including proxy responses, has been examined by independent researchers and consistently produced intraclass correlation coefficients between 0.86 and 0.99.^{13 - 15} The sensitivity and responsiveness of the FIM instrument have also been investigated with excellent results.¹⁶ A summary of the national data collected by the UDSMR is published annually and provides benchmarks for inpatient medical rehabilitation.^{17 - 18} Deaths during the inpatient rehabilitation stay are not included but represent a very small (0.2% or less) mortality rate.

In developing the prospective payment system for inpatient medical rehabilitation,^{4 - 5} CMS extensively reviewed the UDSMR data and associated information collection protocols.¹⁹ The review found that UDSMR hospitals included a large portion of the Medicare rehabilitation cases from most states¹⁹ and that patient demographics, hospital characteristics, and resources used by Medicare beneficiaries were well represented by the UDSMR database.¹⁹ Information regarding the representativeness of the UDSMR database is included in technical reports published by the CMS.^{4 ,19 - 20}

This study was reviewed and approved by the University of Texas Medical Branch Institutional Review Board. Because of the nature of the study, the board waived the need for patient informed consent.

Admission, discharge, and follow-up data were reviewed for 226 137 patients receiving inpatient medical rehabilitation services from 1994 through 2001. The data were collected from 744 hospitals in 48 US states (no data from Alaska or Hawaii). We refined the sample by including patients from the 5 CMS impairment groups in the UDSMR database with the greatest number of patients: orthopedic conditions (30%), stroke (26%), brain dysfunction (7%), spinal cord dysfunction (traumatic and nontraumatic) (5%), and other neurologic conditions (5%). These 5 impairment groups include patients (n = 162 819) from the rehabilitation impairment groups developed by the UDSMR and adopted by the CMS to create the rehabilitation impairment categories for prospective payment (Table 1).

We used clinical criteria developed in previous research on case-mix groups to exclude patients whose rehabilitation was atypical.^{21 - 22} We excluded patients with missing or out of range values including logarithm LOS 3 standard deviations or more above the mean for the impairment group (n = 4577), and cases with incorrect rehabilitation impairment categories or ICD-9 codes (n = 641). In addition, we excluded patients who were younger than 16 years (n = 1602), were readmitted or transferred from another rehabilitation facility (n = 6011), or were admitted for evaluation only (n = 1181). The remaining 148 807 patients comprised the sample and represented 91% of the usable patient records from the original sample for the 5 impairment groups. Patients transferred to acute care facilities (<0.1%) generally do not have discharge FIM scores and so are not included in this analysis.

Not all facilities contributing data to the UDSMR collect follow-up information using the National Follow-up Services. The percentage of cases with follow-up data in the UDSMR database ranged from 34% to 42% for 1994 through 2001. Comparisons between the demographics and FIM instrument data for patient records with follow-up data vs those without follow-up data revealed no important differences. In the full database, 60.2% of the cases were female (59.2% in the sample with follow-up), 83% were non-Hispanic white (81% in the sample with follow-up), mean age was 66.3 (SD, 14.6) years (67.8 [SD, 15.8] years in the sample with follow-up), and median LOS was 20 days (interquartile range [IQR], 13-27) (median [IQR], 20 [13-29] in the sample with follow-up).

Length of Stay. Length of stay was calculated as the total number of inpatient rehabilitation days. When a patient was transferred to an acute care hospital and returned to the initial rehabilitation service within 30 days, we counted only those days the patient was in the rehabilitation service.

Effectiveness. Information on effectiveness was obtained by subtracting the patient’s admission score on the measure of functional status from his/her discharge score.

Efficiency. Efficiency was defined as the change in functional status from admission to discharge divided by the LOS. The shorter the LOS for a given change in FIM rating, the higher the efficiency rating.

Living Setting. Information on living setting was collected at both discharge from medical rehabilitation and at follow-up (80 to 180 days after discharge). Living setting in the UDSMR database is coded as home, board and care, intermediate care, skilled nursing facility, hospital, rehabilitation facility, and other. These categories were collapsed to home and not home for statistical analyses.

Mortality. Information on death after discharge was recorded at follow-up through interviews by nurses from the National Follow-up Services with a family member or caregiver.

These measures have been collected, analyzed, and reported annually as national benchmarks for the rehabilitation community since 1990.^{16 - 17}

We used descriptive statistics to examine differences in demographic characteristics, living setting, effectiveness, and efficiency. Analysis of covariance with repeated measures (time) was used to examine changes in LOS and days to follow-up. Quasi-likelihood analyses were used to assess outcomes from 1994 to 2001.^{23 - 26} The models were longitudinal with time (1994 to 2001) and admission, discharge, and follow-up as independent variables. Dependent variables were effectiveness, efficiency, follow-up living setting, and mortality.

Covariates included age (recorded at admission), marital status (single, married, widowed, separated/divorced), sex, race/ethnicity (coded by patient report as non-Hispanic white or other), living situation (included alone, with family/relatives, or other), payment source (coded as Medicare/Medicaid or other), and comorbidities (other diseases based on ICD-9-CM and total number of comorbidities coded as 0, 1-3, or >3). Ethnicity/race information was included as a covariate because previous research has reported differences in functional status outcomes based on ethnicity/race. Time from disease/injury onset to rehabilitation admission and time from discharge to follow-up were included as time-varying covariates. In some analyses, LOS, living setting at discharge, and admission (baseline) FIM instrument scores were also included as covariates.^{27 - 29}

Separate analyses were computed for each impairment group. When statistically significant effects were found, we conducted post hoc analyses using orthogonal comparisons for trend.^{23 ,28} We used an overdispersed Poisson quasi-likelihood model.²³ In all models a log linear time trend was found to be adequate. P<.05 was considered significant and adjustments were made for multiple comparisons by a modified Bonferroni correction.²⁹ All statistical tests were conducted using SAS version 8.0 (SAS Institute Inc, Cary, NC) or SPSS version 12.0 (SPSS Inc, Chicago, Ill) software. Details regarding the quasi-likelihood approach are provided by Zeger and Liang.³⁰

The final sample included 148 807 records for patients from 744 facilities who received inpatient medical rehabilitation from 1994 through 2001. Descriptive and demographic information for all patients by impairment group is shown in Table 2. Table 3 presents FIM instrument effectiveness and efficiency scores, discharge and follow-up setting, and mortality by impairment group. Table 4 contains admission, discharge, and follow-up functional status information for each year of the study by impairment group. Table 5 contains descriptive longitudinal information on outcomes and covariates.

The total LOS collapsed across all impairment groups decreased from a median of 20 days (IQR, 13-29) in 1994 to 12 days (IQR, 7-19) in 2001 (F = 67.8, P<.001). The proportional decrease in median LOS was greatest for patients with orthopedic conditions (42%). Table 6 provides LOS, effectiveness, and efficiency information over the 8 years. Separate quasi-likelihood models were computed for each of the 5 impairment groups. Effectiveness declined slightly, but this change was not believed to be clinically significant. The largest improvement in efficiency was found for orthopedic conditions (F = 91.4, P<.001), for which the efficiency index improved from 1.6 (SD, 1.0) in 1994 to 2.9 (SD, 1.8) in 2001. The smallest change in efficiency was for stroke (F = 66.5, P<.001), for which the efficiency index improved from 1.2 (SD, 0.9) in 1994 to 1.7 (SD, 1.4) in 2001.

There has not been a substantial change in the percentage of patients discharged to home or living at home following inpatient medical rehabilitation from 1994 to 2001. The range of percentages for persons discharged home after rehabilitation was 79% to 82% for stroke; 81% to 88% for brain injury; 85% to 90% for neurologic conditions; 85% to 92% for spinal cord injury; and 88% to 92% for orthopedic conditions. The percentages living at home at follow-up also remained stable (81%-85% for stroke; 84%-88% for brain injury; 84%-88% for neurologic conditions; 88%-93% for spinal cord injury; and 91%-94% for orthopedic conditions). Patients with orthopedic conditions were most likely to be discharged home and/or living at home at follow-up (>90%). Persons with stroke were the least likely to be discharged home (mean, 81%) or living at home at follow-up (mean, 83%).

Changes in mortality over time are presented in the Figure. The increase in mortality was small but consistent across the 5 impairment groups. Overall mortality from discharge to follow-up (80 to 180 days) increased from a mean of less than 1% in 1994 to 4.7% in 2001. Days to follow-up remained stable (Table 5). We initially assumed that the increase in mortality might be related to sicker patients with more comorbidities being referred to rehabilitation in 2001 vs 1994. However, examination of FIM admission scores revealed little change across impairment groups from 1994 (mean [SD], 72.5 [21.1]) to 2001 (mean [SD], 72.4 [18.7]). We also evaluated age and comorbidities in the quasi-likelihood analyses, hypothesizing that if older patients were admitted more frequently in recent years, mortality rates would increase. However, the average admission age was 69.9 (SD, 15.3) years in 1994 and 67.8 (SD, 15.6) years in 2001. There was no clinically important increase in admission age or number of comorbidities over the study period for any of the 5 impairment groups.

Comorbidities and age were statistically significant covariates in the quasi-likelihood models. Length of stay and admission FIM scores were statistically significant covariates in models in which they were included (living setting at follow-up and mortality). The trend in mortality was statistically significant for all impairment groups in both adjusted and unadjusted models but least dramatic for orthopedic conditions (Figure).

We found that the LOS for inpatient medical rehabilitation decreased significantly from 1994 to 2001 for the 5 impairment groups. No clinically significant change in daily living skills such as dressing and bathing was seen, despite a significant reduction in LOS. Functional status remained relatively stable despite an increase in rehabilitation efficiency. Higher FIM scores are associated with fewer minutes per day of help required from another person to complete basic daily living tasks.^{16 ,31 - 33} Granger and colleagues¹⁶ found that in persons with stroke, each 1-point decrease in FIM score was associated with approximately 2.2 more minutes of assistance by another person to complete activities of daily living. Carter et al⁶ recently reported that an average increase of 1 FIM point from admission to discharge was associated with a 3% reduction in the expected cost of inpatient rehabilitation care. The cost implications varied slightly by functional status item.⁶

The increase in rehabilitation efficiency may be related to changes in the intensity of rehabilitation services. The number of therapy hours provided might not have changed from 1994 to 2001, but the total hours became compressed by providing evening and weekend services.³⁴ Another potential explanation is the pattern of patient recovery. In the early 1990s, patients could have reached a maximum level of functional status early in their rehabilitation program but were allowed to stay in the facility to be sure they would maintain this functional level prior to discharge. This pattern may have changed in the late 1990s when pressures for earlier discharge increased.³⁴ Additional research is needed to determine how the number of rehabilitation therapy hours changed as LOS decreased.

The lack of improvement in effectiveness (absolute change in FIM scores) may be due to a ceiling effect or plateau at which patients have attained their maximum functional status given their residual impairment. This speculation is supported by research from other countries where rehabilitation LOS is much longer than in the United States but discharge FIM scores are similar.³⁵ Another possible explanation for increased efficiency is that patients are beginning rehabilitation earlier. The time from disease/injury onset to rehabilitation admission was shorter in 2001 (median [IQR], 5 [3-10] days) than in 1994 (median [IQR], 8 [6-16] days) due to the decrease in acute care hospital LOS. There is evidence that earlier admission to rehabilitation produces improved functional outcomes for some impairment groups.³⁶

Rehabilitation professionals have also made advances in promoting clinical research and adopting the principles of evidence-based practice.³⁷ The entry-level qualification for many rehabilitation professions is now at the master’s or doctoral level. Empirical evidence regarding the impact of these changes on patient outcomes is lacking, but the increase in professional standards, enhanced research, and use of evidence-based models of care, along with advances in surgical and acute care, may partially account for the improvement in rehabilitation efficiency. These are areas in need of additional research.

A plausible explanation for increased mortality is more difficult to pinpoint. The increase could be due to patients with less severe disabilities being discharged to home health or subacute rehabilitation units in nursing homes, thus leaving a larger proportion of patients with stroke and other severe disorders receiving inpatient medical rehabilitation.

For example, clinical pathways are now well developed for some diagnostic groups, including patients with lower extremity joint replacement or hip fracture. The current trend based on these pathways is to discharge persons with some conditions (eg, lower extremity joint replacements) to subacute rehabilitation units in nursing homes or to home health care without a period of traditional inpatient rehabilitation.³⁸ However, the proportion of patients in different impairment groups remained constant over the study period.^{17 - 18} In addition, the FIM admission scores and number of comorbidities remained stable, suggesting that changes in discharge patterns are not directly related to increased mortality in this sample.

In previous research using the UDSMR database, we reported increased hospital readmission rates associated with decreasing LOS for persons receiving inpatient medical rehabilitation.³⁹ The percentage of persons rehospitalized across 8 impairment groups from discharge to follow-up (80-180 days) increased from 14.8% in 1994 to 18% in 1998. This change in rehospitalization is consistent with our finding of increased mortality. We also examined LOS for persons who died (median [IQR], 15 [9-22] days) compared with persons who did not die (median [IQR], 14 [9-23] days) for the impairment groups included in this study. The difference in LOS was small and not important.

The decrease in LOS for acute care hospitals means that patients entered rehabilitation sooner in 2001 than in 1994 (Table 5), and this might contribute to increased mortality. That is, patients who previously would have died in acute care were transferred or discharged to inpatient rehabilitation. We do not believe this is an adequate explanation because patients must be medically stable before they are sent to an inpatient rehabilitation service. The potential interaction, however, between decreased LOS in acute hospitals and rehabilitation facilities is complex and should be explored.

As LOS in acute care hospitals continues to decline, the use of postacute services is increasing.^{40 - 42} From 1994 to 2000, the number of Medicare patients transferred from acute care hospitals to inpatient medical rehabilitation facilities and skilled nursing settings increased by 22% and 19%, respectively.⁴² The majority of these patients had orthopedic impairments or stroke. The percentage of patients in the UDSMR database with orthopedic impairments ranged from 33% to 36% from 1994 to 2000. The percentage of patients with stroke ranged from 26% to 30% for this period.^{17 - 18} It is unknown what the most appropriate postacute care setting is for patients with stroke and other impairments. A recent review notes that research on whether postacute care services are equivalent has been inconclusive.⁴²

The limitations of this investigation include those associated with analyzing a large database.⁴³ The sociodemographic information in the UDSMR database is obtained from existing medical records and self-reports. While the consistency of the information collection process has been extensively examined,^{6 ,19 - 22} the possibility of coding and reporting errors exists. Another limitation is the lack of information regarding acute care hospitalization. The UDSMR database includes detailed patient information that begins when a person enters inpatient rehabilitation. A priority for our future research is linking the UDSMR database with Medicare files containing information on acute care. Access to this information will help better define the sample, allow comparisons with cases not in the UDSMR database, and provide medical history and background that may help explain the increase in mortality.

While the UDSMR database has been found to be representative of Medicare patients receiving inpatient rehabilitation across the United States,^{19 - 20} it is not a complete record of all rehabilitation facilities nationally and the representativeness for non-Medicare patients is unknown.

Determining the causes of the increase in rehabilitation efficiency and mortality requires further study. Our goal was to document the recent change in LOS and examine its association with functional status, living setting, and mortality. This goal is important in view of the introduction of a prospective payment system for inpatient medical rehabilitation by CMS in January 2002.⁴⁴ Our findings provide a baseline with which to compare future LOS, effectiveness, efficiency, mortality, and other outcomes important to health care professionals, researchers, and consumers to help evaluate how change in LOS influences patient care and outcomes.