Are We Selecting the Right Patients for Stroke Rehabilitation in Nursing Homes?

Article Outline

• Abstract
• Methods
  • Setting and participants
  • Data
  • Intervention
  • Outcomes
  • Propensity for rehabilitation
  • Analysis
• Results
• Discussion
• Conclusions
• References
• Copyright

Abstract 

Murray PK, Dawson NV, Thomas CL, Cebul RD. Are we selecting the right patients for stroke rehabilitation in nursing homes?

Objective

To examine the effect of stroke rehabilitation in the nursing home on community discharge rates and functional status among patients stratified by propensity to receive rehabilitation.

Design

Retrospective cohort.

Setting

Medicaid-certified nursing homes (N=945) in Ohio.

Participants

Patients with stroke (N=2013) admitted to an Ohio nursing home.

Intervention

Rehabilitation therapy services.

Main outcome measures

The propensity to receive rehabilitation, used to adjust for selection bias, was calculated for each patient by using a logistic regression model. Community discharge and change in functional status, measured by using a crosswalk to the FIM instrument, were determined 3 months after admission.

Results

By 3 months after admission, 36.9% of the patients were discharged to the community, 16.6% had died, and 46.5% remained in the nursing home. The overall effect of rehabilitation on community discharge (relative risk [RR]=1.58; 95% confidence interval [CI], 1.33–1.85) was not homogeneous across subgroups stratified by propensity to receive rehabilitation. Patients less likely to receive rehabilitation, as measured by a lower propensity score, had a significant benefit in terms of community discharge (RR=1.65; 95% CI, 1.35–1.97), but those more likely to receive services did not (RR=1.21; 95% CI, 0.87–1.56). Among long-term nursing home residents, rehabilitation services were not associated with improved functional status.

Conclusions

With respect to community discharge, patients who were less likely to receive rehabilitation therapy appear to receive greater benefit from rehabilitation services than those who were more likely to receive rehabilitation. This finding raises concerns about current selection practices for rehabilitation services. Research is needed to identify the patients most likely to benefit, especially in the present fiscally constrained reimbursement environment.

Key words:  Nursing homes , Outcome assessment (health care) , Rehabilitation , Stroke

IN THE PAST 2 DECADES, changes in financing for health care of the elderly have been associated with dramatic changes in the intensity and in the locations in which care is delivered. The introduction of the prospective payment system (PPS) for acute hospital care in 1984 created an incentive for reduced lengths of hospital stay that was accompanied by progressive declines in lengths of stay (LOSs) for virtually all medical and surgical conditions. Stroke care has not been immune to this incentive. Between 1984 and 1998, hospital LOSs for patients with acute stroke declined 51%, from an average of 12.4 days to 6.1 days, resulting in a dramatic increase in discharges to postacute care facilities for stroke rehabilitation.¹ Not surprisingly, this shift of poststroke care was associated with a redistribution of costs for stroke care to skilled nursing facilities (SNFs), where, until recently, there was an incentive to provide rehabilitative care that was accompanied by increasing use of these labor-intensive services.², ³ The growth of these services was an important component of the rapid increase of Medicare expenditures for skilled nursing care through the 1990s, ultimately leading to the passage of the federal Balanced Budget Amendment (BBA) of 1997. A central feature of the BBA was the creation of a PPS for all postacute care, blunting the incentives for providing rehabilitative services in these settings.

In earlier work, we developed a model by using items collected as part of the nursing home admission assessment, which allowed us to describe, with good levels of discrimination, the likelihood or propensity of each patient to receive rehabilitation. By using this propensity measure to control for selection differences, we described the effectiveness of rehabilitation in improving community discharge among patients in SNFs.⁴ The current investigation extends this work by determining whether the benefits observed in our earlier work were similar across subgroups of stroke patients stratified by their propensity to receive rehabilitative care and whether these benefits extended to changes in functional status among patients who remain in skilled nursing home settings after 3 months.

Back to Article Outline

Methods 

Setting and participants 

Potentially eligible subjects were patients with the diagnosis of stroke admitted for the first time to 1 of 945 Medicaid-certified SNFs in Ohio between March 24, 1994, and September 30, 1996. To reduce the potential for residual cohort bias related to the sequential cross-sectional nature of the data,⁵ we included only patients admitted in the last week of each quarter during this period, as described elsewhere in detail.⁴ For this investigation, we also included only patients admitted from an acute care hospital, excluding 293 patients admitted from home. Based on analyses not reported here, about 5% of the sample is patients who received care in acute rehabilitation units before nursing home admission. In the data used for this study, there is no way to identify such patients.

Data 

Data were obtained from the Ohio Minimum Dataset Plus (MDS+) and Ohio’s death certificate files from 1994 to 1997. The MDS+ is a well-validated assessment tool used to collect a comprehensive description of each patient admitted to Ohio’s Medicaid-certified nursing homes, including demographics, preadmission living arrangements, diagnoses present on admission, and, during the initial week in the nursing home, measures of physical health, functional status, psychosocial well-being, activity preferences, medications, specific treatments, restraints, indicators of quality of life, and the receipt and amount of rehabilitation services provided.⁶ Reliability studies that included the State of Ohio have supported the use of MDS+ across sites, except for the measures related to delirium,⁷, ⁸, ⁹ which are not used in the analyses reported here. The admission MDS+ is completed during the first 2 weeks of the nursing home stay and is repeated at the end of each quarter, allowing longitudinal evaluation of functional status in patients who remain in the nursing home. Unique patient identifiers facilitate linkage of MDS+ data with Ohio’s vital statistics files, to permit identification of subject deaths. The institutional review board at the MetroHealth Medical Center approved this investigation and confidential management of study data.

Intervention 

Receipt of rehabilitation services was defined from MDS+ as receiving more than 45 minutes of a rehabilitation therapy (physical therapy [PT], occupational therapy [OT], or speech therapy [ST]) during the first week of the admission. This cutoff was used because it represented the minimum amount of rehabilitation therapy required for incremental reimbursement under Ohio’s nursing home financing policies during the study period.

Outcomes 

Two outcomes were examined. Discharge to the community was defined when an admitted patient was neither alive in a nursing home, as identified from MDS+ files, nor dead, as identified by Ohio’s death certificates, by the end of the quarter after his/her nursing home admission. Significant functional improvement was defined as a 10-point gain in the motor component of the 100-point functional independence measure, derived from the MDS+,¹⁰ at the end of the quarter after his/her nursing home admission. The 10-point gain was chosen arbitrarily as a level of change that would be of clinical significance. Such a change represents an improvement of slightly less than 1 level of assistance (eg, from moderate to minimal assistance) in the items used to derive the scale.

Propensity for rehabilitation 

We used a modification of a previously described logistic model⁴ to define each patient’s propensity to receive rehabilitation services.¹¹, ¹² The modeling estimates a probability (between 0 and 1) called the propensity for receiving rehabilitation therapy for each patient, based on clinical and social characteristics. The clinical and social characteristics included were 108 patient descriptors in the MDS+ recorded at the time of nursing home admission. The variables included items from the following domains of the MDS+ instrument: demographics, measures of social supports, advanced directives, a cognitive performance scale, communication ability, vision, mood, psychosocial well-being, daily activity level, activities of daily living (ADLs) on nursing home admission, use of assistive devices, nutrition, the presence of specific comorbid conditions, and medical treatments. The propensity score, derived from this process, allows us to identify both patients who are highly likely to receive rehabilitation services but who did not, as well as patients who were unlikely to receive rehabilitation services but who nevertheless did. It allows us to adjust directly for these 108 variables as the outcomes among patients who did and did not receive rehabilitation are compared.

The logistic model had a c statistic of .78, indicating good discrimination for receipt of rehabilitation. Visual evaluation of the graph of the predicted and actual distribution of propensity deciles and the Hosmer-Lemeshow goodness of fit test (χ₈² test=6.2, P=.63) indicated that the model calibration was good.

Analysis 

By using logistic regression, we estimated the effect of rehabilitation on each outcome after adjusting for propensity score and measures of severity of illness. Severity covariates included age, cognitive score, ADL score, bowel and bladder continence, body mass index (BMI), nonoral methods of feeding, number of medications, presence of comorbid illnesses, medical treatments, premorbid living arrangements, and evidence of visits from family. Odds ratios were converted to risk ratios by using the technique of Zhang and Yu.¹³

We then divided the sample into quintiles of propensity (ie, 5 equal-sized groups based on propensity for receiving rehabilitation services) and examined a plot of the effect of rehabilitation. We tested the homogeneity of effect across these 5 strata by using the technique described by Woolf.¹⁴ Based on these results, we conducted the remainder of the analyses with the sample divided into 2 subgroups: patients with lower propensity for rehabilitation (the lowest 3 quintiles) and patients with higher propensity for rehabilitation (the highest 2 quintiles).

We repeated the logistic regression analyses in the 2 redefined propensity strata, to determine the association of rehabilitation with both community discharge and functional improvement, controlling for the clinical severity measures. Finally, we compared the sociodemographic and clinical characteristics of the 2 propensity subgroups by using chi-square and Wilcoxon signed-rank tests. All statistical analyses were conducted by using SAS, version 8.2.^a

Back to Article Outline

Results 

The sample consisted of 2013 patients admitted from a hospital to a nursing home after a stroke. Of these 2013 patients, 1442 (72%) received rehabilitation and 571 (28%) did not. By 3 months after nursing home admission, 743 (36.9%) patients were discharged to the community, 335 (16.6%) had died, and 935 (46.5%) remained in the nursing home. Of the 935 who remained in a nursing home, 885 (95%) had motor ADL scores available both at admission and 3 months after admission (fig 1).

• 
  • View Large Image
  • Download to PowerPoint

• Fig 1. 

  Patient flow in the first 3 months after admission to the nursing home.

Figure 2 displays the effect of rehabilitation on community discharge for the overall sample and by quintile of propensity. Overall, after adjusting for patient propensity for rehabilitation and severity of illness, patients who received rehabilitation were 1.58 (95% confidence interval [CI], 1.33–1.85) times more likely to be discharged to the community than those who did not. By quintile, this beneficial effect is only significant in the 3 quintiles least likely to receive rehabilitation. Not surprisingly, the relative rates of community discharge were not homogeneous across the quintiles (χ₄² test=4.91, P=.30). After dichotomizing the quintiles into lower and higher propensity subgroups, the relative rate of community discharge associated with the use of rehabilitation in the lower-propensity subgroup was 1.65 (95% CI, 1.35–1.97), whereas the relative rate in the higher-propensity subgroup was 1.21 (95% CI, 0.87–1.56).

• 
  • View Large Image
  • Download to PowerPoint

• Fig 2. 

  Relative rates of community discharge in the sample, divided into quintile of propensity. Horizontal bars represent the 95% confidence intervals.

Table 1 compares the baseline characteristics of the total study sample and the 2 subgroups of patients with lower and higher propensity for rehabilitation. Patients less likely to receive rehabilitation were slightly older and more likely to be white. Patients less likely to receive rehabilitation had a lower prevalence of depression, more cognitive deficits, and more problems with bowel and bladder continence; they also were more likely to have do-not-resuscitate (DNR) orders (all P<.005). Of note, patients less likely to receive rehabilitation also were much more likely to have Medicaid as their primary insurance, and, although they were less likely to have lived alone before admission, they were less likely to have regular contact with relatives during their nursing home stays (all P<.005).

Table 1. Baseline Characteristics of Study Sample and Propensity Subgroups

Baseline Characteristic	Propensity for Rehabilitation		Total	Level of Significance
	Lower	Higher
No. of subjects	1183	830	2013
Age ± SD (y)	78.4±10.2	77.4±10.2	78.0±10.2	.02
Sex (% female)	60.1	60.0	60.1	.96
Race (% white)	88.5	85.0	87.0	.02
BMI ± SD (kg/m2)	23.8±4.9	24.7±5.3	24.2±5.1	<.001
Motor ADL score ± SD⁎	45.3±21.3	46.5±17.5	45.8±19.8	.03
Cognitive impairment score ± SD†	3.1±2.1	2.2±1.8	2.7±2.0	<.001
Lived alone (%)	27.6	36.3	31.2	<.001
Regular contact with relatives (%)	80.2	85.5	82.4	.002
Bladder continent (%)	27.7	34.2	30.4	<.001
Bowel continent (%)	38.7	47.4	41.4	<.001
Diabetes (%)	31.2	30.4	30.8	.75(NS)
Depression (%)	13.4	18.1	15.3	.004
CHF (%)	21.7	24.5	22.9	.15
DNR status	44.5	25.9	36.8	<.001
Medicaid insurance	16.7	11.3	14.5	.001
Hours of rehabilitation per week	7.2	9.2	8.2	<.001

Abbreviations: CHF, chronic heart failure; DNR, do not resuscitate; NS, not significant; SD, standard deviation.

Of the 885 patients who had motor ADL measurements both on admission and after 3 months, 125 (14.1%) showed improvements in scores of 10 points or greater. Few baseline characteristics were significantly associated with improvements on bivariate analysis, and these all tended to reflect poorer baseline condition. Patients more likely to show functional improvements had lower baseline motor ADL scores (39.3 among patients who did improve vs 43.7 among patients who did not improve, respectively, P=.05), lower likelihood of bladder continence (14.8% vs 24.2%, respectively, P=.04), and a higher likelihood of congestive heart failure (27.2% vs 19.6%, respectively, P=.05). Although patients with improved motor ADL scores were more likely to have received rehabilitation services (72.0% vs 65.9%, respectively), this association was not significant on bivariate analysis (P=.18). In the multivariable analysis, neither the effect on the overall remaining sample (relative risk=1.05; 95% CI, 0.64–1.71) nor the effect stratified by propensity for rehabilitation showed an association of rehabilitation with improved motor ADLs.

Back to Article Outline

Discussion 

In the United States, as the LOS in the acute hospitals after stroke has declined, Medicare-supported nursing home admissions after stroke have increased almost 3-fold between 1987 and 1998, from 36,063 to 103,913.¹⁵ In 1999, 25% of all Medicare patients hospitalized for acute stroke were discharged to nursing homes, with most receiving rehabilitation services for the purpose of improving functional independence and increasing the likelihood of eventual return to the community. The financial disincentive for providing such services that is integral to the BBA of 1997 makes it imperative to identify more clearly those stroke patients who are most likely to benefit. In the absence of data from controlled trials of rehabilitation in the nursing home setting, patient selection likely will be guided by local practice patterns and forces that may be unrelated to evidence of benefit.

Although our earlier work showed the overall effectiveness of rehabilitation for improving rates of community discharge,⁴ the results of this investigation suggest that there are patient subgroups who benefit more than others. We found that patients who were less likely to receive rehabilitation therapy on admission actually had higher relative rates of community discharge at 3 months when they received rehabilitation, whereas those who were more likely to have received rehabilitation did not. From a clinical perspective, these lower-propensity patients were more likely to be cognitively impaired and incontinent of bowel and bladder function and to have somewhat lower motor ADL scores on admission. These findings support those of Kane et al,¹⁶ who reported better results for poststroke rehabilitation among patients who were sicker on admission.

Our results also suggest that lower propensity for rehabilitation was associated with factors that are not strictly clinical in nature, including being insured by Medicaid, having less regular contact with relatives, and having DNR orders written on admission. Although similar results regarding the selection process have been reported by others,¹⁷, ¹⁸ there is no evidence that these characteristics are more or less likely to be associated with beneficial effects of rehabilitation.

There are 2 plausible explanations why patients with a higher propensity to receive rehabilitation services apparently did not benefit in terms of community discharge. First, higher-propensity patients may have been more likely a priori to be discharged to the community, both for clinical and for nonclinical reasons. The results displayed in table 1 suggest that these patients may have been less disabled on admission and hence more likely to be discharged home, regardless of the provision of rehabilitative services. Because the MDS+ data system does not record standard measures of stroke severity, such as the National Institutes of Health Stroke Scale,¹⁹ it is not possible to determine from this study whether the higher-propensity patients had less severe strokes by conventional measures. By similar argument, it may be that, although patients with problems of cognition and continence had better relative improvement with rehabilitation, the higher absolute levels of baseline cognition and continence among higher propensity patients (see table 1) may have made their transition home less challenging to caregivers. With regard to other measures of a social nature, it is noteworthy that patients with higher propensities for rehabilitation were significantly more likely to have regular contact with their relatives, substantially lower likelihood of DNR status on admission, and greater likelihood of medical insurance other than Medicaid. The presence of good social support has been described elsewhere¹⁷, ¹⁸ as being associated with the selection of patients with stroke to receive rehabilitation. Second, among those more likely to receive rehabilitation, we may have failed to detect a true beneficial effect from rehabilitation because of inadequate statistical power. Of the patients in the top 2 propensity quintiles, more than 90% (750/830) actually received rehabilitation, limiting our ability to detect a beneficial effect.

Almost half the admitted stroke patients remained in a nursing home after 3 months, and these patients appeared not to benefit from rehabilitation in terms of improved functional status. On bivariate analyses, only lower baseline ADL scores and higher levels of bladder incontinence and prevalence of congestive heart failure were significantly associated with an improvement in functional status. Of note, the first 2 of these also were associated with a lower likelihood of rehabilitation (see table 1) and a greater benefit from rehabilitation in terms of community discharge.

As an effectiveness analysis,²⁰ the results of this investigation may understate the benefits of an “ideal” program of PT, OT, and ST among stroke patients admitted from an acute care hospital to a nursing home. Indeed, the purpose of our study was to examine the effectiveness of rehabilitation as provided in more typical settings, adjusting in the most rigorous way possible for differences in patient severity and selection for treatment. Rehabilitation services in our study were defined by data used for administrative as well as clinical purposes across 945 Medicaid-certified nursing homes, including facilities in urban and rural areas, those affiliated and unaffiliated with academic medical centers, and for-profit as well as not-for-profit ownership. Each of these characteristics may be associated with different types and intensities of rehabilitative services.²¹ Thus, among patients who are less likely to receive rehabilitation, the benefit of rehabilitation on community discharge described in this study is likely to underestimate the effect of an ideal program. In addition, among higher-propensity patients, our findings of no effect and of no effect on functional status among patients who remain in the nursing home may understate the true effect of an ideal program of rehabilitation. Although it may be tempting to adjust statistically for the differences in nursing home characteristics, to do so would be to lose most of the between-nursing home variation. It is exactly this variation that is the natural experiment we are exploiting in this observational study.

Because of the BBA of 1997-related disincentive for providing rehabilitation in nursing home settings, it has become increasingly important to produce hard evidence on which to base decisions about which stroke patients should receive such services and what types and intensity of services should be provided. In the absence of such evidence, rehabilitation therapy for stroke may be discouraged generally, or its provision may be directed disproportionately to those who are least likely to benefit. The consequence of such decisions include not only the failure to optimize stroke patients’ outcomes but also a paradoxical increase in government-sponsored costs for skilled nursing and long-term care.

Back to Article Outline

Conclusions 

With respect to community discharge, patients who were less likely to receive rehabilitation therapy appear to receive greater benefit from rehabilitation services than those who were more likely to receive rehabilitation. Rehabilitation services do not seem to be successful in improving function for patients who become long-term residents. From a clinical perspective, these findings should cause rehabilitation professionals to reexamine the decision process about who should receive rehabilitation in the nursing home setting. Until research helps to better define who should be treated, a more liberal approach to selecting patients for rehabilitation should be pursued—that is, trials of rehabilitation therapy for more impaired patients should be encouraged. Clinical research should be designed to better characterize who receives benefits from these services and why functional improvement is not better among long-term nursing home residents. Researchers also need to examine how the changes that have been prompted by the 1997 BBA have influenced treatment and outcomes among patients with stroke.

Supplier

Back to Article Outline

References 

1. Health Care Financing Administration . Health Care Financ Rev Stat Suppl . 2000;23:44
   • View In Article
2. Murray PK , Singer ME , Fortinsky R , Russo L , Cebul RD . Rapid growth of rehabilitation services in traditional community-based nursing homes . Arch Phys Med Rehabil . 1999;80:372–378
   • View In Article
   • Abstract
   • Full-Text PDF (865 KB)
   • CrossRef
3. Health Care Financing Administration . Health Care Financ Rev Stat Suppl . 1999;22:29
   • View In Article
4. Murray PK , Singer M , Dawson NV , Thomas CL , Cebul RD . Outcomes of rehabilitation services in nursing home residents . Arch Phys Med Rehabil . 2003;84:1129–1136
   • View In Article
   • Abstract
   • Full Text
   • Full-Text PDF (90 KB)
   • CrossRef
5. Detsky AS , O’Rourke K , Corey PN , Johnston N , Fenton S , Jeejeebhoy KN . The hazards of using active clinic patients as a source of subjects for clinical studies . J Gen Intern Med . 1988;3:260–266
   • View In Article
   • MEDLINE
   • CrossRef
6. In:  Feldman J editors. Minimum Data Set + training manual . Natick: Eliot Pr; 1992;
   • View In Article
7. Gruber-Baldini AL , Zimmerman SI , Mortimore E , Magaziner J . The validity of the minimum data set in measuring the cognitive impairment of persons admitted to nursing homes . J Am Geriatr Soc . 2000;48:1601–1606
   • View In Article
   • MEDLINE
8. Hawes C , Morris JN , Phillips CD , Mor V , Fries BE , Nonemaker S . Reliability estimates for the Minimum Data Set for nursing home resident assessment and care screening (MDS) . Gerontologist . 1995;35:172–178
   • View In Article
   • MEDLINE
   • CrossRef
9. Mor V . A comprehensive clinical assessment tool to inform policy and practice (applications of the minimum data set) . Med Care . 2004;42(4 Suppl):III50–III59
   • View In Article
   • MEDLINE
10. Williams BC , Li Y , Fries BE , Warren RL . Predicting patient scores between the functional independence measure and the minimum data set (development and performance of a FIM-MDS “crosswalk.”) . Arch Phys Med Rehabil . 1997;78:48–54
    • View In Article
    • Abstract
    • Full-Text PDF (1017 KB)
    • CrossRef
11. D’Agostino RB . Propensity score methods for bias reduction in the comparison of a treatment to a non-randomized control group . Stat Med . 1998;17:2265–2281
    • View In Article
    • MEDLINE
    • CrossRef
12. Rosenbaum PR , Rubin DB . Reducing bias in observational studies subclassification on the propensity score . J Am Stat Assoc . 1984;79:516–524
    • View In Article
13. Zhang J , Yu KF . What’s the relative risk? A method of correcting the odds ratio in cohort studies of common outcomes . JAMA . 1998;280:1690–1691
    • View In Article
    • MEDLINE
    • CrossRef
14. Greenland S , Rothman KJ . Introduction to stratified analysis . In:  Rothman KJ ,  Greenland S editor. Modern epidemiology . Philadelphia: Lippincott-Raven; 1998;p. 253–279
    • View In Article
15. Health Care Financing Administration . Health Care Financ Rev Stat Suppl . 1999;22:44
    • View In Article
16. Kane RL , Chen Q , Blewett LA , Sangl J . Do rehabilitative nursing homes improve the outcomes of care? . J Am Geriatr Soc . 1996;44:545–554
    • View In Article
    • MEDLINE
17. Gibson CJ . Epidemiology and patterns of care of stroke patients . Arch Phys Med Rehabil . 1974;55:398–403
    • View In Article
    • MEDLINE
18. Unsworth CA . Selection for rehabilitation (acute care discharge patterns for stroke and orthopaedic patients) . Int J Rehabil Res . 2001;24:103–114
    • View In Article
    • MEDLINE
19. Brott T , Adams HP , Olinger CP , et al.   Measurements of acute cerebral infarction (a clinical examination scale) . Stroke . 1989;20:864–870
    • View In Article
    • MEDLINE
    • CrossRef
20. Hearst N , Grady D , Barron HV , Kerlikowske  . Research existing data . In:  Hulley SB ,  Cummings SR ,  Browner WS ,  Grady D ,  Hearst N ,  Newman TB editor. Designing clinical research . Philadelphia: Lippincott Williams & Wilkins; 2001;p. 195–210
    • View In Article
21. Changes in new admissions to Ohio nursing homes following the implementation of the RUG-III Reimbursement System: 1994–5. MEDTAPP Program. Columbus: Ohio Dept of Human Services; Sept 1998.
    • View In Article
22. Morris JN , Fries BE , Mehr DR , et al.   MDS Cognitive Performance Scale . J Gerontol . 1994;49:M174–M182
    • View In Article
    • MEDLINE

• a SAS Institute Inc, 100 SAS Campus Dr, Cary, NC 27513.