Direct Evidence

There is no direct evidence from randomized trials about the effect of early neurological rehabilitation on health outcomes.

Indirect Evidence

Conclusions for Question 1

One small, retrospective, observational study from a single rehabilitation facility supports an association between the acute institution of formalized, multidisciplinary, physiatrist-driven TBI rehabilitation and decreased length of stay (acute hospital and acute rehabilitation) and some measures of short-term physiologic (noncognitive) patient outcomes (MacKay, Bernstein, Chapman, et al., 1992). The level of evidence is Class III. This study concerned patients with severe brain injury (GCS 3-8); there is no evidence from comparative studies for or against early rehabilitation in patients with mild or moderate injury.

Some indirect evidence also confirms that early rehabilitation is associated with a shorter inpatient rehabilitation LOS, but this association rests on important assumptions that have not been examined in prospective studies. Most important, the association of acute LOS with rehabilitation LOS and greater rehabilitation costs does not directly imply that shortening acute LOS will result in favorable changes in these outcomes. A common confounding variable in the studies reviewed in this section is the inability to control for the possible correlation between the velocity of recovery and the acute hospital LOS. Patients with TBI who have similar GCS scores on admission may recover at markedly different rates. If a patient evidencing rapid recovery reaches threshold for rehabilitation admission and continues to recover quickly thereafter, his or her acute rehabilitation LOS and total LOS days will be fewer than someone who reaches the same landmark at a slower rate. Because present indicators of TBI severity do not measure rate of recovery, this oversight might explain why the relation between acute LOS and rehabilitation LOS persists even after statistical control for severity of illness on admission.

Another finding in the studies reviewed in this section is that rehabilitation admission FIM and acute care LOS are strongly associated with rehabilitation LOS and outcome. This finding suggests that acute hospitalization LOS is not simply a proxy for injury severity or level of recovery on transfer to rehabilitation.

Future Research on Question 1

In essence, this question addresses the efficacy of starting formal rehabilitation efforts very early during the acute-care stay at the trauma center as opposed to "filling in" until the patient is transferred to a rehabilitation program. Certain therapeutic modalities such as physical therapy are generally felt to be properly started soon after admission because of the known rapidity with which complications such as contractures begin. Although the details of the proper techniques, timing, and intensity of such treatment remain to be determined, it is unlikely that a control group of patients that would receive no acute physical therapy could be ethically formed. Although the indications for early application of other rehabilitation-oriented therapeutic modalities are less clear, similar ethical constraints will likely prevent the development of pure control groups for any specific discipline.

The ability to study the efficacy of a formalized program, however, is not subject to such constraints at the present time. The concept of the acute initiation of formal rehabilitation should be defined as an attempt to begin rehabilitation independently of the patient's location or other extraneous constraints such as medical complications, bed availability, and so forth. Such a formalized system is an attempt to "blur the line" between the stay at the trauma center and the time at a rehabilitation center. It would attempt to divorce the treatment from the milieu so that the patient's needs would drive the treatment.

As such, the integration of such a formalized program into an acute care center's operating procedures could be effected in a prospectively randomized fashion without ethical constraints. Since such rehabilitation efforts properly come under the aegis of a physiatrist, the primary independent variable would be the involvement of a physiatrist overseeing explicit formalized application of rehabilitation techniques to the "experimental" group as compared with continuation of the status quo in the "control group." The dependent variables would be acute care and rehabilitation LOS, outcome at time of discharge from acute care (admission to inpatient rehabilitation), outcome at discharge from rehabilitation, and cost-effectiveness of resource utilization.

For such an investigation to work, patients would have to be classified into working categories at a very early stage so that formalized and standardized rehabilitation protocols designed to meet their needs could be applied. If every patient were to be treated differently, it would not be possible to control for the resulting confounding of treatment variables with the independent variable. On the other hand, since different patients need a different array of therapeutic modalities (differing in terms of treatments, timing, and intensity), managing all patients in the same fashion would not be proper. Therefore, if the research is to be useful, such issues must be addressed prior to onset of the investigation.

It also would be necessary to strictly define the applied therapies, since these would be confounding variables in the analysis. Issues such as timing, intensity, modalities, therapist training, milieu, and so forth would have to be standardized within and between treatment groups. This will be especially critical if multiple centers are to be included in the study and combined in the data analysis.

Such a study would not address which modalities should be applied at what point during the acute care stage. These are separate questions addressing the efficacy of rehabilitation modalities in general. The suggested investigation would, however, address the present, seemingly artificial dependence of the initiation of formal rehabilitation on extraneous variables which commonly occur during the early postinjury period.

Challenges in Assessing the Effectiveness of Inpatient Rehabilitation

A precise knowledge of the natural, untreated prognosis of brain injury could reduce uncertainty about the effectiveness of inpatient rehabilitation, but such knowledge is lacking. Because experimental trials of inpatient rehabilitation are unlikely to be performed, investigators have relied on statistical methods to adjust for differences in the baseline characteristics between the groups of patients compared in studies. These groups might be patients who received different intensities of rehabilitation services, received rehabilitation services relatively early or late, or did not receive rehabilitation in the usual course of care. The validity of these methods depends in large part on the predictive ability of the risk factors measured in these studies.

As discussed below, the likelihood of a good outcome depends on many patient characteristics. For this reason, it is impossible to interpret studies that fail to describe the baseline characteristics of the sample under study. In such studies, it is not clear whether the results were due to the interventions under study or to unreported selection factors. Two extremes characterize these studies with respect to the description of populations and samples. On one hand, it is common to have a sample described as "patients who were considered ready for (the intervention) by their occupational therapists" or "consecutive referrals to a vocational rehabilitation program who were considered employable under the right circumstances." On the other hand, the inclusion criteria for the sample may be a lengthy list of narrow parameters, including scores falling within a specific range on a series of neuropsychological tests. In one case, there may be no description of the patients. In the other, the description may be so specific that the results do not apply to the greater proportion of patients.

The nature of rehabilitation makes it difficult to evaluate its effectiveness. Multidisciplinary rehabilitation is a complex intervention. Even in studies that provide evidence that patients undergoing rehabilitation improved, it usually is not possible to determine which specific components of rehabilitation are effective. In general, little description of the precise components of multidisciplinary rehabilitation programs is available. Some studies use the number of hours of performance of individual treatment modalities (e.g., physical therapy, occupational therapy, speech therapy, etc.) as a measure of the intensity of rehabilitation. However, additional hours of specific treatments may be provided to patients who enter rehabilitation with more severe deficits. In addition, their control for the confounding variables that they collected would have been considerably strengthened by the use of multivariate statistical methods such as regression analysis. Even without this confounding, an aggregate measure like time spent with the patient cannot capture the social factors and relationships that can be important components of the therapeutic process.

At present there is no reliable method to measure the effect of exposure to the milieu of the rehabilitation program--the interactions between patients and other patients, nurses, therapists, and physiatrists during the course of an inpatient TBI rehabilitation stay--or to separate their effects from the content of the actual therapy provided. Such information may be critical when attempting to determine whether an inpatient rehabilitation unit or a skilled nursing facility may be interchangeable for a given patient.

Most studies do not provide even descriptive information about the components of rehabilitation and the content of specific interventions. In these studies, rehabilitation is somewhat of a black box--it is defined, by default, as whatever happens between admission to and discharge from a rehabilitation unit. The lack of detail about what constitutes rehabilitation reduces the generalizability of each study's findings and makes it difficult to compare the results of different studies attempting to assess the effectiveness of rehabilitation.

In formulating a strategy for reviewing the literature, we focused on whether information was available to examine the actual mechanisms by which inpatient TBI rehabilitation affects outcomes. Specifically, we sought to examine whether the results of rehabilitation vary according to (1) whether the intervention was directed and managed by a physiatrist and (2) the number, kinds, and frequency of methods applied. Secondarily, we sought to examine which factors predict a good outcome and how these factors may be used in decisions about how and when patients might benefit from inpatient rehabilitation.

The population for this question consists of people who sustained TBI between the ages of 18 and 65 years whose injury severity warranted a trip to a hospital emergency department, transfer to acute care, and subsequent transfer to inpatient rehabilitation. We also intended to focus predominantly on studies that included or measured the following patient characteristics:

• Age.
• Glasgow Coma Scale score.
• Severity of injury.
• Multiple injuries.
• Premorbid data.
• Mechanism of injury (kind of trauma).
• Intracranial diagnosis.
• Functional status.

Finally, studies had to report one or more of the following outcome measures:

• Length of stay in a rehabilitation facility.
• Immediate care costs and long-term financial burden.
• Health status at discharge from inpatient rehabilitation.
• Long-term measure of impairment.
• Long-term measure of disability.
• Independence, relationships, family life, satisfaction.

Of the 87 papers included for review of questions 1 and 2 (see Figure 3), 57 had some relevance to question 2. Of these, 10 primarily addressed predictors of outcome, 22 were uncontrolled followup studies of inpatient rehabilitation, and 20 examined the usefulness or validity of various measures of outcomes. Five studies, which were controlled or quasiexperimental studies that addressed the effectiveness or intensity of inpatient rehabilitation, are discussed in detail in the following sections.

How effective is acute inpatient TBI rehabilitation in general?

Before addressing whether the intensity of inpatient TBI rehabilitation is associated with improved outcome, we examined the more general question of the effectiveness of TBI rehabilitation itself. A large number of uncontrolled case series show that people with brain injuries generally improve by the time of discharge from the acute inpatient rehabilitation facility (Ashley, Persel, and Krych, 1993; Basso, Capitani, and Vignolo, 1979; Ben-Yishay, Silver, Piasetsky, et al., 1987; Cope, Cole, Hall, et al., 1991; Cope and Hall, 1982; David, Enderby, and Bainton, 1982; Dresser, Meirwosky, Weiss, et al., 1973; Eames and Wood, 1985; Evans and Ruff, 1992; Johnston, 1991; Jones and Evans, 1992; Lomas and Kertesz, 1978; Malec, Smigielski, DePompolo, et al., 1993; 1Mills, Nesbeda, Katz, et al., 1992; Panikoff, 1983; Pazzaglia, Frank, Frank, et al., 1975; Prigatano, Fordyce, Zeiner, et al., 1984; Sarno, 1976; Scherzer, 1986; Tuel, Presty, Meythaler, et al., 1992). Because of imperfect knowledge about the natural history of TBI and the nearly complete absence of data about the results of alternative methods of rehabilitation after discharge from the acute care hospital, these Class III studies provide only weak evidence that inpatient rehabilitation is effective. Comparing these studies and aggregating their results into a systematic examination of results was not possible because data were too incomplete to discern the relationship between types of populations or interventions and outcomes.

One quasiexperimental study used an unmatched control group to assess the effectiveness of acute inpatient TBI rehabilitation (Aronow, 1987). Sixty-eight patients were selected from 107 consecutively discharged patients treated at a single inpatient brain injury rehabilitation center. Their long-term outcomes were compared with those of 61 patients selected from 1,400 cases consecutively entered into an epidemiologic database of TBI inpatients at a neurosurgical unit in an area of the country with no comprehensive rehabilitation available for severe TBI. These two groups were termed "rehabilitation" and "nonrehabilitation," respectively. The selection criteria were TBI (> 1 hour of unconsciousness and > 24 hours of altered consciousness), age at injury between 5 and 80, acute hospital LOS > 15 days, and not comatose at the time of acute hospital discharge.

Measures of TBI severity in the Aronow (1987) study were PTA, acute hospital LOS, presence or absence of open brain injury, and number of skull fractures. Age, sex, race, and years postinjury were measured as confounding variables. TBI severity and the other potentially confounding variables were controlled for by using regression analysis, entering the confounding factors into the model prior to adding the rehabilitation versus no rehabilitation variable. The outcome measure was a 13-variable measure that included vocational status; living arrangement; number of recent inpatient treatment episodes; number of recent outpatient episodes; hours of daytime care required; functional status in self-care, mobility, and residential skills; number of home and outside social contacts; and number of physical, cognitive, and emotional symptoms. This standardized outcome measurement was developed unique to this study and has not been otherwise tested. Outcome data were obtained via telephone interview with the person with TBI or a caregiver/relative during a set study period not indexed to time after injury or rehabilitation. Chi square analysis was used to examine differences in PTA between groups, and linear multiple regression modeling was used to control for confounding variables in determining the relationship between rehabilitation and outcome.

At baseline, the rehabilitation and nonrehabilitation groups differed significantly in PTA, the major index of TBI severity used in the study (Aronow, 1987). Seventy percent of the rehabilitation group had PTAs > 4 months, while 74 percent of the nonrehabilitation group had PTAs 1 month. The nonrehabilitation group also was less impaired in self-care activities and memory.

In a multiple regression model adjusting for age, sex, race, injury severity (PTA, acute hospital LOS, open brain injury, number of skull fractures), and years postinjury, rehabilitation was associated with a better long-term outcome (Aronow, 1987). The overall R2 value was 0.551, suggesting that about one-half of the variance in this group was accounted for by the nine predictors plus rehabilitation. Days in acute hospital, duration of posttraumatic amnesia, age at onset, sex, and whether rehabilitation was performed were statistically significant predictors of outcome. However, the correlation coefficient (Pearson r) for rehabilitation was only 0.159, suggesting that only about 3 percent of the variance was related to whether or not rehabilitation was done.

This study (Aronow, 1987) mildly supports the hypothesis that acute inpatient TBI rehabilitation improves outcome. The finding of a benefit despite worse initial severity in the rehabilitation group lends some credence to the results. The study has important weaknesses that have been enumerated by others (High, Boake, and Lehmkuhl, 1995). The obvious baseline differences between the two groups means that the attempt to identify a suitably comparable control population failed. While the statistical analysis was well done, this method of control works best when there is good reason to believe that the two groups being compared are similar. The differences also reflect the underlying problem that the subset of patients admitted to a rehabilitation unit are not representative of the population thought to benefit from it.

Because this was a retrospective study (Aronow, 1987), the authors were limited to information that had been recorded in the patients' charts or (in the case of the control group) data recorded in an epidemiologic study, although all records were abstracted using the same protocol and all followup interviews were conducted using an identical instrument and process. Data on the timing of followup (how long after injury, acute hospital discharge, and rehabilitation discharge the outcome data were collected) were not available. GCS data also were unavailable. Finally, the use of a proprietary outcome instrument prevents comparison of their data to other studies.

Is the intensity of acute inpatient TBI rehabilitation services related to outcome?

There are no prospective randomized controlled trials of different levels of intensity of acute rehabilitation. Four observational studies, three of which are retrospective, addressed the relationship between the intensity of rehabilitation services and outcomes for people with brain injury not due to stroke (Aronow, 1987; Blackerby, 1990; Heinemann, Hamilton, Linacre, et al., 1995; Spivack, Spettell, Ellis, et al., 1992) (see Evidence Table 2).

A retrospective multicenter study of 140 patients admitted between 1990 and 1991 to one of eight rehabilitation hospitals was the best of these studies (Heinemann, Hamilton, Linacre, et al., 1995). Although the study was retrospective, all of the participating hospitals were prospectively collecting data using the Uniform Data Set for Medical Rehabilitation (Granger, Hamilton, and Sherwin, 1986). Intensity of therapy was defined either by the number of billed hours of individual therapeutic modalities (physical therapy, occupational therapy, speech and language services, and psychological services) or by all services combined. The authors examined whether a higher level of services was associated with better motor and cognitive FIM scores, achievement of motor or cognitive potential ([D/C FIM-admit FIM]/[100-admit FIM]), and efficiency of change ([D/C FIM-admit FIM]) at the time of discharge.

An analysis of the interrelationships between intensity and severity of injury or other descriptors revealed that they were not independent. Intensity of treatment covaried with functional status at admission, patient demographics, and medical characteristics. This suggests that the functional status on admission is actually related to the therapy intensity the patient receives. It appears that the patients received more therapy if they were admitted with less cognitive function, had uninterrupted stays, had a longer delay to admission, were younger, and so forth.

Investigation of the relationship between intensity of therapy and their (Heinemann, Hamilton, Linacre, et al., 1995) various outcome measures (discharge motor and cognitive FIM scores, achievement of motor or cognitive potential, and efficiency of change) did not reveal any significant relationship for occupational, physical, or speech therapy intensities. There was also no significant intensity:outcome relationship for intensity of all therapies combined. Only the number of hours of psychologic work per day, usually delivered as cognitive therapy, were associated with any alterations in outcome. These alterations were improvements in discharge cognitive, FIM score, achieved potential gains in cognitive FIM score, and efficiency of cognitive recovery.

The major weaknesses of this paper (Heinemann, Hamilton, Linacre, et al., 1995) are the absence of a specific definition of TBI and lack of control for severity of injury as a confounding, predictive variable. It is unlikely that the admission FIM will cover all of the variance otherwise subsumed by GCS, PTA, and/or duration of unconsciousness (coma). There was no control over or description of differences in treatment between the involved hospitals. Also, the use of billing hours as the index of therapeutic intensity probably included time not spent directly in patient care. Finally, the authors only used one outcome measure (FIM), and there is no long-term followup. Despite these weaknesses, however, the use of prospective data collection and credible analytic techniques make this the most important paper to address the issue of the relationship between intensity of therapy and outcome. It is the only Class II study in this category.

A retrospective study (Spivack, Spettel, Ellis, et al., 1992) examined the influence on outcome measured at rehabilitation discharge of therapeutic intensity during the first treatment month and over the entire stay on 95 patients with TBI. The cohort consisted of patients with a complete set of records who had been admitted to a single inpatient rehabilitation unit between 1988 and 1990. A definition of TBI was not given. It was noted that not all patients were comatose on admission. LOS ranged from 20 to 412 days with a median of 58 days.

Intensity of treatment was calculated as hours of actual treatment performance measured in 15-minute intervals for PT, OT, ST, cognitive remediation, vocational services, neurophysiology, respiratory therapy, therapeutic recreation, and medical services. Intensity of treatment during the first month was the total hours of treatment during that month. Subjects were separated into high- and low-intensity groups based on the median split of treatment hours during the first month. The median was 76 hours with a range of 18 to 196 hours. Average daily intensity of treatment over the entire stay was calculated, and subjects were again classified into high- and low-intensity groups based on the median split. The median was 4 hours per day with a range of 1.4-12.25 hours. The authors (Spivack, Spettel, Ellis, et al., 1992) felt that the true time spent per weekday was probably about one-third higher, since these estimates did not take account of days when therapy could not be administered (passes, holidays, weekends, etc.).

GCS was measured within 24 hours of admission to the trauma hospital. Head AIS score, duration of coma, severity of extracranial injuries (highest non-head AIS), and time since TBI were also measured. The statistical method used to control for these confounding variables was unclear.

The independent variables were intensity of treatment during the first month of rehabilitation, average daily intensity of treatment over the entire LOS, and LOS. All of these independent variables were made binary using the median split method as described above.

The dependent variables were outcome measures. RLA scores were measured on admission and discharge. In addition, therapy-specific outcomes on admission and at discharge were assessed using a seven-point functional status scale developed by clinicians within each rehabilitation discipline. A principal components analysis was used with a varimax rotation conducted on the matrix of correlations among functional scale scores at admission to group the scores on various axes. This resulted in grouping along axes of physical performance, higher-level cognitive skills, and cognitively mediated physical skills. In addition, patients were rated on their RLA scores on admission and discharge.

The statistical methods for assessing the relationships between the independent and dependent variables were analyses of variance and covariance controlling for multiple comparisons.

Analysis of covariants (ANCOVA) with repeated measures analysis was used to investigate treatment intensity and LOS on the dependent variables of admission and discharge scores on physical performance, higher level cognitive skills, cognitively mediated physical skills, and RLA level. In this analysis, LOS and intensity of treatment during the first month of rehabilitation and LOS and average daily intensity of treatment over the entire LOS were separately analyzed.

LOS significantly influenced outcome across all outcome groups. With respect to either intensity of treatment during the first month of rehabilitation or average daily intensity of treatment over the entire LOS, the only statistically significant relationship was between discharge RLA and 1-month treatment intensity. Spivak, Spettel, Ellis, and colleagues (1992) found a borderline nonsignificant relationship (p=0.06) between higher level cognitive skills and average daily treatment intensity. They also found a borderline nonsignificant relationship (p=0.07) for the triple interaction of RLA, LOS, and average daily treatment intensity. Based on this borderline relationship, the authors performed univariate ANOVA analysis on this interaction. This revealed a significant effect of high-intensity treatment during the entire stay on RLA scores on patients with longer stays. This relationship did not hold for patients with short lengths of stay. ANOVA analyses of age and LOS as confounding variables suggested that these variables could not explain the correlation.

There are a number of weaknesses in the paper by Spivak, Spettel, Ellis, and colleagues (1992). Because a precise definition of TBI was not provided, it is difficult to determine the parent population. Additionally, the focus of the analyses was on outcome measures that were derived in the unit and were, therefore, of unestablished validity and reliability. The major weaknesses in this study, however, are the use of the median split method to dichotomize the independent variables and the lack of powerful multivariate statistical methods.

In this case, the median split method is a dichotomizing method of convenience and does not necessarily reflect any underlying physiologic basis. The distribution of intensity times may be influenced by confounding influences of various origins, including brain and extracranial injury characteristics, patient personality, payer characteristics, and so forth. One method to force independent variables into binary distributions would be to determine split thresholds recursively in terms of their influence on outcome. Alternatively, intensity of treatment could have been analyzed during the first month of rehabilitation and average daily intensity of treatment over the entire LOS as continuous variables. The use of the median split method in dividing independent variables might have increased the likelihood of a type II error.

The other major weakness in the paper by Spivak, Spettel, Ellis, and colleagues (1992) is the lack of powerful, multivariate control for confounding variables. As was demonstrated (Heinemann, Hamilton, Linacre, et al., 1995), it is not proper to assume independence between intensity of therapy and severity of injury or other patient descriptors. The analysis would have been considerably strengthened by using regression-type analysis.

In the analysis of results, several inferences were made from the nonsignificant but borderline interactions between higher level cognitive skills and average daily treatment intensity (p=0.06) and the triple interaction of RLA, LOS, and average daily treatment intensity (p=0.07). Based on the latter borderline relationship, a significant effect was found of intensity of treatment during the entire stay on RLA scores on patients with long LOSs. Based on this interaction, suggestions were made on managing the intensity of treatment for patients with more severe injuries. However, the original interaction that inspired these suggestions was not significant (p=.07); consequently, the value of the suggestions is not substantial.

In one retrospective study (Blackerby, 1990), the influence on brain injury outcome of a change in mean daily therapeutic intensity that accompanied a major programmatic change at the study institution was investigated. The study took place at two commercial inpatient head injury rehabilitation provider units run by Rebound, Inc., a commercial provider of head injury rehabilitation services. The charts of all 149 patients with brain injury in the program between 1986 and 1988 were evaluated; 97 percent of the patients were admitted with a diagnosis of TBI. There was no description of TBI provided. Patients were either in a coma treatment program or an acute treatment program. For the prechange group, 55 percent were in the coma treatment program (54 of 98 patients), whereas only 27 percent of patients in the postchange group were in the coma treatment program (14 of 51 patients).

There was no precise measure of TBI severity. Confounding variables quoted in this study included age, level of function on admission, and length of time postinjury as measured on admission. The measure of function on admission was not specified. It was reported that the two groups did not differ with respect to these variables, although the method of handling them as confounding variables is not stated and the raw data are not provided.

The independent variable was intensity of therapy measured as mean number of daily therapy hours for all types of therapy combined. The two groups were formed in 1986 when the rehabilitation provider units altered the structure of their rehabilitation service delivery system. At this time, the intensity of inpatient rehabilitation was increased from an average of 5.5 hours per day to an average of 8 hours per day.

The dependent variable was inpatient rehabilitation LOS. The relationship between the independent and dependent variables was analyzed using t statistics, separately analyzing the coma treatment and acute treatment groups.

The results demonstrated a large change in average length of stay in both the coma and acute treatment programs following the programmatic changes. The variability in the LOS also decreased after the programmatic change. The only statistical examination was a t-test between pre- and postchange LOS for both the coma treatment program and the acute treatment program. Both of these changes were statistically significant as tested.

The differences between these groups in terms of cost was evaluated. The average daily cost for the rehabilitation programs was $785 per day with $350 representing the fixed costs. For the patients in the coma treatment program, the average savings would be $16,950 per patient. For the patients in the acute treatment program, the average savings would be $18,504. It was noted that such cost savings, as well as the decreased variability in LOS that appeared to accompany the change in mean daily therapeutic intensity, would be of use to insurance carriers in predicting and controlling costs.

There are a number of weaknesses in the paper by Blackerby (1990). A strict definition of TBI was not provided, making it difficult to determine the parent population. In addition, there was a lack of control for severity of TBI or other confounding variables. No data on these variables were provided.

The major weakness, however, was the lack of statistical controls for a number of potentially significant confounding variables that are intrinsic to this experimental design. It appears there were major changes in this rehabilitation delivery system that accompanied the increase in mean daily therapy intensity. The paradigm change was described as a change to a "naturalistic activity, total therapeutic day model" that apparently involved adaptations of those activities of interest to the individual before head injury. It was suggested that implementation of internal case management was part of the new system and that "senior clinical staff were added to the programs in the roles of clinical consultants and case managers, which increased staff experience and personnel." The occurrence of programmatic changes of such magnitude will, almost by definition, alter patient management in ways outside of those resulting from the increase in mean daily therapy intensity. If all of these changes could be described and quantified, their confounding influences could be addressed using multivariate statistics. In the absence of these data and such statistical analyses, it is difficult to interpret the results of this report.

Conclusions for Question 2

Based on the current literature, there appears to be little evidence that therapeutic intensity, measured as hours of treatment, is related to the beneficial effects of acute, inpatient TBI rehabilitation when the analysis controls for confounding variables. The only Class II study (Heinemann, Hamilton, Linacre, et al., 1995) found no correlation between intensity of individual or grouped therapeutic interventions and outcome. The second study found statistically significant correlations only for discharge RLA and 1-month treatment intensity (Spivack, Spettell, Ellis, et al., 1992). Nonsignificant trends were reported toward associations between higher level cognitive skills and average daily treatment intensity and for a triple interaction of RLA, LOS, and average daily treatment intensity, but the interpretation of these trends is unclear in the absence of statistical significance or other supporting evidence. The third report (Blackerby, 1990) appears to have been so highly confounded by uncontrolled variables as to render questionable any comparative interpretation of findings.

There are a number of possible reasons why various intensities of rehabilitation do not appear to correlate with functional improvements. First, the effect of specific comorbidities was underinvestigated in these papers. Second, there is a lack of long-term followup in all three studies. Third, these studies did not examine the quality of treatments or the reasons the various therapies were applied.

Another potential explanation for the demonstrated lack of correlation between therapeutic intensity and outcome is that all patients were receiving enough therapy and that added hours did not make a difference. Similarly, the ranges of intensities of treatment (lack of treatment variability) may have been too limited to show differential effect. As Heinemann, Hamilton, Linacre, and others (1995) noted, the Health Care Financing Administration (HCFA) mandated 3 hours per day of therapy for each patient starting in 1983. The legislation may have decreased practice variation that, prior to the regulations, might have been wide enough to affect patient outcomes. Future studies should either consider suspending such constraints or including the influence of such mandated decreases in variation in therapeutic effort into the power calculations used to determine the minimal size of their patient populations.

Overall, however, it also is debatable whether hours of applied therapy is the proper index for therapeutic intensity. The impact of individual therapeutic disciplines may not be independent or even separable, and the time spent in each might not be the best index of their intensity.

The use of hours of applied therapy as the index for therapeutic intensity also raises the question of how to measure the "milieu effect" of comprehensive rehabilitation. The potential contributions to recovery that might arise from formal and/or informal patient-patient, patient-nurse, patient-therapist, and patient-TBI rehabilitation environments have not been addressed in any study to date. Particularly in units devoted to TBI, such a milieu effect should be taken into account in attempting to determine the mechanism of efficacy of the present rehabilitation efforts. This is particularly relevant to such questions as whether delivery of rehabilitation services to a similar group in a setting outside of a formal inpatient rehabilitation unit (i.e., a less expensive setting) is an equally efficacious and, therefore, acceptable alternative method of care delivery.

Future Research on Question 2

Future research into the question of intensity of inpatient rehabilitation must deal specifically with the limitations highlighted in the present body of literature. These deal specifically with the way the question has been asked and generally with the details of describing the patient population and the therapies applied.

The present unidimensional definition of intensity as hours of application appears to be neither an appropriate definition of intensity nor an adequate descriptor of the therapies. In order to examine the importance of hours of application, there must be a description of and control for the other aspects that make up each therapy. These include modalities used, therapist training, interactions between therapeutic disciplines, and so forth, as well as the milieu in which the therapies are delivered. For instance, it is questionable if it is valid to compare two separate physical therapy sessions solely in terms of time spent without addressing what is done within those sessions, who performs the therapies, which aspects of other treatment modalities (for example, cognitive therapy) might be imbedded, etc. Such confounding variables either need to be standardized (preferable) or described in a fashion amenable to subsequent statistical control.

It also is necessary to better describe the patient population being treated. It is highly unlikely that all people with TBI will receive optimal benefit from the same general therapeutic approach. It is critical that the types and magnitudes of impairments resulting from the TBI be described for the patient population, including both the severity of injury and the resultant degrees of physical and cognitive dysfunction. If adequate descriptions are provided, it will be possible to determine the interaction of the various facets of the individual treatment modalities with the types of impairment demonstrated by the people being studied. In addition, it will facilitate subsequent focused studies addressing matching treatment protocols to patient subtypes.

If treatments can be standardized and the patient population can be adequately described, it is possible that RCTs could be performed addressing hours of therapy as the independent variable and outcome as the dependent variable. With the proper standardization, the influence of general milieu also could be addressed by adding it as a second independent variable. Such investigations, if performed in fashions that are replicable and comparable between studies, should prove extremely valuable in furthering our understanding of optimizing types and intensities of treatments for people with specific, defined levels of TBI-induced impairments.

Direct Evidence

Indirect Evidence

Observational Research

Although research designs without control groups are limited, they can be a source of hypotheses which could be tested in controlled trial settings. This section highlights insights from studies with uncontrolled research designs identified in our literature search.

Nine observational studies of clients before and after cognitive rehabilitation fulfilled the criteria for inclusion in this review (Cicerone and Giacino, 1992; Deacon and Campbell, 1991; Glisky, Schacter, and Tulving, 1986; Goldstein, McCue, Turner, et al., 1988; Middleton, Lambert, and Seggar, 1991; Ponsford and Kinsella, 1988; Ruff, Mahaffey, Engel, et al., 1994; Scherzer, 1986; Wilson, Evans, Emslie, et al., 1997). One study used a measure of a relevant health outcome, everyday memory failures (EMFs), to evaluate treatment effect, and it is presented in Evidence Table 8 (Wilson, Evans, Emslie, et al., 1997).

The other eight studies either compared clients' performance from baseline phase to treatment phase, provided the same or similar treatments to different matched groups, or combined group and individual methods of measurement. In general, results indicate that for the selected clients treated in these clinical studies, one-on-one interaction with therapists in a rehabilitation environment is likely to improve individual performance on targeted laboratory tasks. Because the studies are not comparative, the improvement observed does not contribute to the body of evidence about the intervention being provided. However, the fact that clients do in fact improve gives rise to innovations in rehabilitation technology that may be useful to people with TBI and thus warrant further evaluation.

For example, in the study presented in Evidence Table 8 (Wilson, Evans, Emslie, et al., 1997) 15 clients were provided an electronic device, programmed to assist them in remembering to do routine daily tasks. Prior to the intervention, they were interviewed to identify targets for memory remediation unique and important to them. Thus the intervention was individually adapted. The score for everyday memory failures (EMFs) was the number of times a target was forgotten. EMFs were measured for 2 to 6 weeks during baseline. During the treatment phase, which lasted 12 weeks, each person in the study wore and used the device. The return-to-baseline phase was 3 weeks.

All participants had significant decreases in EMFs during treatment. During return-to-baseline, EMFs increased for 11 of the 15 participants; five increases were statistically significant. The results of this study suggest that the use of an electronic cueing device decreases EMFs for some people with TBI. These findings also contribute to the evidence for the link represented by Arc 1 of the causal pathway. The observational design of this study weakens its value as evidence of effectiveness. However, in considering that the nature of most of the interventions reviewed here are not individually adapted and on face value do not appear to be as pragmatic as an effective reminder device, this study is useful in that it generates a hypothesis about an intervention that may have the potential to prosthetically improve memory for a person with TBI.

Conclusions for Question 3

Very few controlled studies of cognitive rehabilitation have examined health outcomes or employment. Two small randomized controlled trials (Class I) and one observational study provide evidence of the direct effect of compensatory cognitive devices (notebooks, wristwatch alarms, programmed reminder devices) on the reduction of EMFs for people with TBI. A second randomized controlled trial provides evidence that compensatory cognitive rehabilitation reduces anxiety and improves self-concept and interpersonal relationships for people with TBI.

In the absence of strong and sufficient evidence for a direct effect of cognitive interventions on health and employment, we examined a causal pathway linking cognitive rehabilitation to intermediate measures of cognition and subsequent associations between those measures and health or employment. Two small randomized controlled trials (Class I) provide limited evidence that practice and CACR improve performance on laboratory-based measures of immediate recall. However, no studies evaluated the link between cognitive tests and health outcomes, and associations between performance on cognitive tests and posttrauma employment and productivity were inconsistent.

Future Research on Question 3

Identifying and evaluating outcomes that are relevant to people with TBI and their families is the first priority of a research agenda. Among the studies we reviewed, perhaps the most pragmatic outcome measure used was that of everyday memory failures. It is possible that the absence of treatment effect in these studies could be a function of the study's lack of relevance in the lives of the people being evaluated, represented in outcome measures and interventions that have little meaning to those people.

It is also important to identify the laboratory tests that are strongest and most reliable in their ability to measure cognitive function in relevant contexts, and to standardize their use across research projects and hospital and clinical settings.

Two other questions for future research on topics not specifically addressed in this review are: (1) When is the client ready for the intervention? and (2) What are the markers of that readiness? Large, multicenter comparisons may provide initial information for a research design to investigate these questions.

In general, the studies in this review that did not produce a treatment effect compared one form of cognitive rehabilitation with another form, CACR to non-CACR practice, and were specific to unstructured rehabilitation methods. Treatment effects were not observed when one kind of remediation was compared with another when there were equal levels of stimulation for both treatment and comparison groups. What are the differential effects of general stimulation and technology? As TBI rehabilitation technology grows, costs escalate. Consequently, certain subsets of the total population of survivors--those with liberal insurance policies and/or private money--will receive the intervention. It is important for clients as well as payers to know if the new technology leads to improvement or whether the increased level of stimulation used to deliver the technology results in the improvement.

Definitions

There are at least five models of supported employment: (1) individual placements, (2) work enclaves, (3) apprenticeships, (4) small businesses, and (5) mobile work crews (Powell, Pancsofar, Steere, et al., 1991). The most common model is individual placement, which provides training and ongoing support individually to each survivor, in settings where fewer than 5 percent of the workers are disabled. The aim is to provide a quality match between worker and job requirements, including high job satisfaction by both worker and employer and decent wages (Ellerd and Moore, 1992). This is the model usually recommended for survivors of TBI (Wehman, Kreutzer, Stonnington, et al., 1988), though a variation of the apprenticeship model has also been tried (Curl, Fraser, Cook, et al., 1996). One of the most detailed definitions of the individual placement model is by Kreutzer, Wehman, Morton, and colleagues (1988). They identify four essential components, all of which must be tuned to the type of deficit of the client (in this case, the client with TBI):

• Job placement. This includes (a) matching job needs to client abilities and potential, (b) facilitating employer and client communications, (c) facilitating caretaker communications, (d) arranging travel arrangements or training, and (e) analyzing the job environment to detect potential problems.
• Job site training and advocacy. This emphasizes the active role of the employment specialist, job coordinator, or job coach, who is often cited as the key professional in supported employment programs (Wehman, West, Fry, et al., 1989; Ellerd and Moore, 1992). The job coach serves functions usually left to the employer in conventional vocational rehabilitation (e.g., training). The job coach also is proactive in identifying problems and designing solutions in cooperation with all the parties involved.
• Ongoing assessment. This involves continuous monitoring of key aspects of the client's work performance. There is an intense intervention by the job coach at the beginning, but this is expected to diminish, or "fade," as the client settles into a successful work adjustment (Ellerd and Moore, 1992; Wehman, Sherron, Kregel, et al., 1993). This process is well illustrated in Figure 1 of the article by Wehman, Kreutzer, West, et al., 1990.
• Job retention and follow-along. A continuing, proactive process in which the job coach tries to anticipate problems and intervene early to prevent crises from disrupting the client's adapted job placement. This assistance to the client is of indefinite duration, although it is expected to diminish over time (Wehman, Sherron, Kregel, et al., 1993).

The hallmark of supported employment methods is that they are applied on the job, in the actual work environment, to help the client succeed. Off-job training and practice to prepare the client for work may be an important prelude to supported employment in some programs, but the job coach always accompanies the client to the job site to work out on-the-spot solutions to problems as they arise and to mediate between employer and client. This problem-solving-in-situ is a defining principle of the method of supported employment (Kreutzer, Wehman, Morton, et al., 1988). Finally, these programs usually aim at competitive employment, usually defined as employment in a setting about 95 percent occupied by workers without disabilities and paying at least the official minimum wage (Wehman, Kreutzer, West, et al., 1990; Ellerd and Moore, 1992). They do not aim at placement in sheltered workshops or other settings designed primarily to accommodate the needs of disabled workers. There is a strong commitment to placing the client in the highest job position possible and to approach or exceed the preinjury level of work.

Supported employment is necessary only in cases where standard vocational rehabilitation is not sufficient to secure the desired level of employment for a survivor of TBI. In this respect, supported employment could be considered an extension of vocational rehabilitation into the actual job site as the final stage of helping a particular survivor resume a productive life. Studies of postinjury employment in which survivors are sorted by severity of injury show that not all survivors of TBI require this extra step. People suffering mild injury (GCS = 13) are usually reemployed at high rates: from about 60 to 85 percent within 1 year postinjury, and this high rate is maintained up to 15 years later (Dikman, Temkin, Machamer, et al., 1994; Schwab, Grafman, Salazar, et al., 1993; Edna and Cappelen, 1987; Fraser, Dikman, McLean, et al., 1988). In those same studies, people suffering from moderate or severe injury do not fare as well. The reemployment rates for moderate injury (GCS = 9 to 12) for the same periods are 50 to 60 percent, and for severe injury (GCS = 8), reemployment ranges from 20 to 30 percent.

The GCS alone may not always be the best predictor of employment success in multivariate analyses including other severity measures (Abrams and Toms Barker, 1991). It seems evident, however, that supported employment is most necessary for severe injury because regular vocational rehabilitation programs are insufficient. Even a variety of other interventions before job placement--such as cognitive training, with and without occupational trials, and behavior modification--show only limited success (Wehman, Kreutzer, West, et al., 1990). In addition, about one-half of survivors with moderate TBI and at least 15 to 20 percent of survivors with mild cases also might benefit from supported employment programs.

Ideally, as an extension of vocational rehabilitation, supported employment would address the problems that lead to lost jobs. Devany and colleagues at the Medical College of Virginia (Devany, Kreutzer, Halberstadt, et al., 1991) found that survivors of TBI who had trouble with reemployment suffered from four kinds of difficulties:

• Somatic problems, like debility, loss of balance, and motor deficiencies.
• Cognitive problems, like loss of memory, inability to focus attention, obsessions, and indecisiveness.
• Behavioral problems, like inertia, restlessness, depression, and impatience.
• Communication and social problems, like contentiousness and various disorders of speech and writing.

They also found that the survivors' scores on the Minnesota Multiphasic Personality Inventory (MMPI) showed peaks on scales 8 (schizophrenia), 4 (psychopathic deviate), and 2 (depression), in that order of magnitude. Valid MMPI scores might be problematic with this population, but the pattern found seems to match the common clinical impression. A configuration of 8-4-2 on the MMPI indicates a depressed person with severe interpersonal and social deficits who could easily alarm or offend others. Add this to the various motor and cognitive deficits already noted, and we have a formidable set of obstacles to placing and maintaining a person with a severe TBI in a job setting. The underlying assumption of the supported employment model is that all people referred to the program are employable under suitable conditions (Wehman, Sherron, Kregel, et al., 1993). Successful programs focus on finding or making those conditions in support of client success at work.

Of the 93 articles retrieved for review, 42 were excluded based on initial exclusion criteria, reducing the total number of articles to 51. Two investigators read all articles retrieved through the database search, as well as five additional articles acquired from reference lists and recommendations from peers, for a total of 56 articles.

Direct Evidence

There is no direct evidence from randomized trials about the efficacy of supported employment.

Indirect Evidence

No Class I or IIa studies of supported employment were found in the literature. In one prospective, controlled, observational study, Haffey and Abrams (1991) measured job placement and retention rates of 130 participants in a program of supported employment, the "Work Reentry Program" or WRP. These results were compared with those of 35 clients in a day-treatment program and 76 individuals who had received no postacute rehabilitation (the "comparison group"). Participants in the WRP program were recontacted every 6 months for up to 3 years. Other subjects were followed for varying amounts of time over 3 years, depending on when they entered the study. A total of 87 (67 percent) of the 130 participants entering the program were placed in employment. Followup for the 87 placements was as follows:

• 1 to 6 months of followup: 18 participants.
• 7 to 12 months of followup: 23 participants.
• 13 to 18 months of followup: 17 participants.
• 19 to 24 months of followup: 15 participants.
• > 24 months of followup: 14 participants.

The second series has been reported in a number of publications (Wehman, West, Fry, et al., 1989; Wehman, Kreutzer, West, et al., 1989; Wehman, Kreutzer, West, et al., 1990; Wehman, Sherron, Kregel, et al., 1993). These studies evaluated the outcomes of supported employment in a prospective registry study. In the first article cited, the sample included five survivors followed for 75 weeks (1.4 years); by the last article in the series the sample had increased to 115 consecutive referrals followed for up to 60 months (5 years). These individuals passed through two consecutive phases of vocational rehabilitation: first, the standard postrehabilitation services, then a supported employment program.

For convenience in the remainder of this report, we will refer to the first study as Haffey and Abrams and to the second series of studies, taken as a group, as the Virginia series, after the home of the researchers in the Medical College of Virginia.

Posttrauma vocational success in the Virginia series was compared with the rates of preinjury employment success of each client, who served as his or her own control in the study. Preinjury employment is thus taken as a baseline control for results in the two subsequent phases of intervention--the last of which was supported employment--as each client passed through the following history: Preinjury employment (baseline) => posttrauma employment (1st phase intervention) => supported employment (2nd phase intervention).

Measures of employment success in the Virginia series were (1) number of jobs held per client, (2) mean hourly wage, (3) mean monthly wage, (4) mean annual wage, (5) monthly employment ratio (MER). The last is an index of vocational success developed by the Virginia group. It is the ratio (with specific definitions of the two variables): number of months client employed/number of months client could have been employed (Wehman, Sherron, Kregel, et al., 1993).

Results were measured over 5 years of operation with measures of outcome taken weekly (Wehman, Kreutzer, West, et al., 1989). During that time, 80 of the 115 entering clients were placed in jobs, but it is not clear how many of the 115 were followed for all 5 years. In the whole series of studies, we have reports on 5 clients in 1989 (Wehman, West, Fry, et al., 1989) and 20 clients later in the same year (Wehman, Kreutzer, West, et al., 1989). In 1990, there is a report of 53 clients (Wehman, Kreutzer, West, et al., 1990) and in 1993 the total is 115 (Wehman, Sherron, Kregel, et al., 1993). From this pattern in the reports, we may infer that about five clients were followed for a full 5 years, 15 clients for between 4 and 5 years, 33 clients for at least 3 years, and 62 for less than 3 years.

Outcomes

Outcomes are summarized in the evidence tables. In the Haffey and Abrams study, 68 percent of the WRP clients secured competitive employment, and only 18 percent were considered chronically unemployed. Only 39 percent of the day-treatment group and 34 percent of the comparison group reported any employment after discharge. At the last followup period reported, 71 percent of the WRP clients placed were still working, but no retention rates for the day-treatment and comparison groups were reported. See Table 7 for a summary of the results from the Virginia studies (Wehman, Sherron, Kregel, et al., 1993).

Table

Table 7. Postinjury employment outcomes of TBI clients in a supported employment program.

Conclusions for Question 4

There is Class IIb evidence that supported employment can improve the vocational outcomes of TBI survivors. Almost all information about supported employment comes from two bodies of work, each of which use different experimental designs and different models of supported employment. The findings have not been replicated in other settings or by other centers, so the generalizability of these programs remains untested.

In the Virginia studies, the clients all had severe cases of TBI as rated by the GCS. In the Haffey and Abrams experiment, no GCS scores are given, but the data provided on severity of injury show median length of coma in the three groups ranging from 6 to 7 days, with the median value for the entire sample of 199 cases at 7 days and the overall range from 0 to > 30 days. In the 80 job-placed survivors of the Virginia series (Wehman, Sherron, Kregel, et al., 1993), the average length of coma was 48 days, and the range was from 0 to 182 days. Clearly, the Haffey and Abrams sample contained many people with moderate or even mild injuries, while the Virginia sample consisted entirely of people with severe injuries. We have already noted, in the Introduction to this report, that severity of injury has an important effect on vocational success.

There also are difficulties in interpretation that are derived from problems with the individual study designs and with confounded variables and biases in the groups being compared. In both studies, the prospective data collection avoids the hazards of retrospective inference, but the lack of a control condition in the Virginia studies and the highly biased allocation of clients to the groups being compared in the Haffey and Abrams research make it impossible to clearly interpret results. In the Haffey and Abrams study, there are so many differences among the groups being compared that comparisons are almost meaningless. The supported employment group is heavily biased toward better vocational outcomes, containing only clients specially selected by vocational counselors as likely prospects for work. The day-treatment group, however, was filled with clients for whom "competitive employment was not a current goal," and the comparison group was made up of clients who had rejected a chance at supported employment and had multiple motives to avoid work altogether. Then, when the three groups were tested for differences in client characteristics (see Table 1, Haffey and Abrams, 1991), the only difference found was that the comparison group was significantly less likely to have been employed at the time of injury. With a stacked deck, no one is surprised when the dealer wins the hand, and the superior vocational performance of the supported employment group in this experiment is likewise an anticlimax.

The problems with the Virginia series are of another sort. They arise from the inherent limitations of the design itself, since case-control studies without separate control groups and unbiased allocation of participants are unable to untangle confounded variables. Repeated measures on a single group are bound to be confounded with variables for which there are no controls. An observation made by Wehman, Sherron, Kregel, et al. (1993) will illustrate the problem. They note that during the economic recession of 1990 to 1992--the worst in the United States since the 1970s--nearly 20 percent of their program participants lost their jobs due to layoffs. The design of the Virginia studies compares baseline levels of employment preinjury with subsequent performance under two levels of vocational rehabilitation, including supported employment as the final stage.

In the Virginia sample, the average age at injury was 24.8 years, and the average age at referral to the program was 30.9 years, making the average interval from injury to entry in the program 6.1 years. The average time to job placement after program entry is about 1 year more, and the subsequent followup in the series of studies under review ranges up to 5 years, with a substantial proportion of the clients at least 3 years into followup. Simple arithmetic discloses that the average time from preinjury employment to postinjury followup is on the order of 10 to 12 years, and for about half the clients the interval will be even longer. This is the time elapsed between baseline measures of preinjury vocational success and postinjury outcome measures under supported employment. It is a significant possibility that economic changes over a 10- to 12-year period (or more, for half of the sample) would have significant effects on employment. Those economic effects will be confounded with the effects of supported employment programs running at the same time, and there is no way to separate the two kinds of effects unless a control group of clients, without supported employment, is measured over the same interval. Likely, there are many other confounded variables at play in the same way, including unknown ones.

In spite of the problems mentioned, it remains true that all the programs of supported employment reviewed showed better rates of vocational success than the baseline expectations of survivors of TBI who received only standard postacute rehabilitation or even with special kinds of preemployment vocational counseling and training (see "Problems Addressed" section in the Introduction to this report). The success documented by the Virginia series with survivors with severe injuries is especially convincing in this respect. Supported employment appears to be a promising way to increase the vocational success of survivors of TBI, but the present literature does not give definitive proof of its effectiveness and does not provide enough clarity on why it works or guidance to the best applications of the method.

Future Research on Question 4

Definitions

Case management has emerged as a possible solution because it systematically monitors client needs over time and facilitates access to services in various institutions and programs across communities. Case managers usually serve people with long-term or chronic conditions because they have complex needs and find it difficult to navigate the health and/or social care systems.

According to the Commission for Case Manager Certification (1996), case management is: A means for achieving client wellness and autonomy through advocacy, communication, education, identification of service resources and service facilitation while ensuring that available resources are being used in a timely and cost-effective manner in order to obtain optimum value for both the client and the reimbursement source (p. 2).

The role typically includes admission or intake assessment, care-plan development, service referral, coordination of service details, and collaboration with care providers, informal caregivers, and the client (Goering, Farkas, Lancee, et al., 1988; Goodwin, 1994) and may also include authorization of service payments (Ashley, Krych, Persel, et al., 1994; Evans and Watke, 1995). The domain is to determine service needs and to access service elements with a sequence and timing that will result in desired outcomes for the client and family as well as desired outcomes such as cost control for the service organization.

Case Management Characteristics and Desired Outcomes

One characteristic of case management is the adopted mission of the employing organization. Since case managers are found in hospitals, rehabilitation programs, health departments, aging services departments, mental health services, insurance companies, and managed care organizations (Gerber, 1994), the focus can vary from acute-care disposition planning to long-term patient advocacy to service cost control. As the orientation and purpose of case management programs vary, outcomes may reflect the differences. In order for case management interventions to be compared and tested for effectiveness, it is important to define the specific purposes and aspects of case management that are provided.

Case managers also focus on helping clients move across institutional or organizational systems and across provider disciplines. Managing these boundary issues calls for a collaborative approach around a common focus--the client and family. Understanding case management role specifications and the extent of boundary work is also critical for comparing programs and interpreting research that tests case management effects.

Within this context of TBI incidence and long-term effects, problems experienced by clients and their families, and the definition and role of case management, our study was conducted to identify evidence of case management effectiveness. The purpose was to review the literature for controlled clinical studies of the influence of care coordination on targeted outcomes among TBI rehabilitation populations of clients and their families.

Of the 69 articles retrieved for review, 27 were excluded based on the initial exclusion criteria, reducing the total number of articles reviewed to 42. Two investigators read all retrieved articles from the database search, as well as relevant articles found on reference lists of the retrieved articles and those obtained from colleagues, for a total of 73 articles. The only criterion for study selection in this phase was that case management was an independent variable. By mutual agreement three studies were critically analyzed and entered into Tables 1 and 2 to report evidence of case management effectiveness in TBI rehabilitation.

Results

Does the provision of long-term care coordination enhance the general functional status of people with TBI? The search strategies yielded three controlled studies of case management effectiveness in TBI rehabilitation. Two studies compared case management with non-case management, and one study compared two types of case management. Two studies were completed trials (Ashley, Krych, Persel, et al., 1994; Greenwood and Brooks, 1994), and one article (Malec, Buffington, Moessner, et al., 1995) presented preliminary results after 1 year of data collection.

Although all the studies addressed case management effectiveness, they differed in most of the design characteristics, as reported in Table 1. Regarding study purpose, Ashley, Krych, Persel, et al. (1994) focused on the level of independence following rehabilitation, while Greenwood and Brooks (1994) addressed the rehabilitation process, client employment and quality of life, and family burden; and Malec and colleagues (Malec, Buffington, Moessner, et al., 1995) measured employment outcomes. Two designs were group comparisons, and one design compared outcome rates with previously established baseline rates (Malec, Buffington, Moessner, et al., 1995). In the only clinical trial that tested the effects of case management on client outcomes (Greenwood and Brooks, 1994), the intervention was allocated to sequentially admitted clients in randomized sites. In the other two studies, sequential admission for eligible subjects was also the intervention allocation method, but the subjects were at different stages of recovery. The number of subject withdrawals and the number of exclusions were reported in two papers; in the third, the subjects were all clients who met the inclusion criteria and could be matched to the control group. The sample sizes were moderately high (> 100) in two studies (Greenwood and Brooks, 1994; Malec, Buffington, Moessner, et al., 1995), and small (n = 39) in one (Ashley, Krych, Persel, et al., 1994).

Each selected study focused on a different subpopulation. In two studies, the subjects were homogeneous for level of disability: either moderate disability (mean Disability Rating Scale [DRS] scores of 4.95 and 5.17) (Ashley, Krych, Persel, et al., 1994); or severe disability (mean DRS scores of 16.2 and 18.3, and mean Glasgow Coma Scale [GCS] scores of 6.6 and 5.5) (Greenwood and Brooks, 1994). In the third study, the sample at 1 year consisted of clients with mild injuries (79 percent), as rated with the GCS and moderate or severe injuries (21 percent) Malec, Buffington, Moessner, et al., 1995). Comparison group differences made it necessary to control for aspects that may have affected the outcomes, but no controls were mentioned in the preliminary report by Malec, Buffington, Moessner, et al., 1995.

The studies also tested different models of case management intervention. Ashley, Krych, Persel, et al. (1994) evaluated two insurance-coverage models in which the key aspects were the authority to approve disability payment and rehabilitation service claims, although this was not defined, and differences between the two groups for this characteristic were not reported. Also, the report did not include any other behavioral or role descriptors to compare and contrast the two intervention models. In fact, although a key independent variable was same versus different case managers, it is not clear from the description whether all subjects in one group had the same case manager or whether each subject in the group had one case manager for the entire post-TBI period. Greenwood and Brooks (1994) tested a medical model of case management, defined as clinical needs assessment during acute hospital care, formulation of a proactive rehabilitation plan, and facilitation of rehabilitation cooperation and involvement among patients, family, and professionals.

Malec, Buffington, Moessner, et al. (1995) evaluated a medical-plus-vocational model of case management. They defined it as assessment and rehabilitation planning during acute hospital care for the provision of medical outpatient rehabilitation services and for vocational counseling and planning related to employment services available in the community. Although there were presumably common role behaviors in these two models, there were not enough details offered in the Greenwood and Brooks (1994) report to determine whether the case manager referrals had led to subsequent care coordination by vocational counselors and others. If so, the Greenwood model would have been similar to the Malec model. Only one study identified the discipline or training that the case managers had received (Malec, Buffington, Moessner, et al., 1995). Also, the intervention periods varied by study and by client need, which no doubt influenced the results.

In addition, other design aspects differed among the studies. The data collection points ranged from 1 month to 24 months, with only two studies measuring outcomes at 12 months. In two projects, the research spanned 2 years, but subjects were followed for less than 2 years if they had not entered the study at the beginning. Also, due to the team-oriented nature of case management and the close institutional quarters of the subjects, blind assessments of the subjects were not possible. Therefore, care delivered by the case managers and other team members could have differed among groups and influenced the outcomes. Additionally, there was almost no information about possible cointerventions--such as other service providers--who may have offered care coordination and continuity, and there was very limited information about the other types of rehabilitation services that may have influenced client outcomes.

There are very few studies of the effectiveness of case management. The results of these studies are mixed (Evidence Table for case management, 2). There is evidence from Class III studies that case management improved vocational status. This was associated with the single case manager and insurance approach (Ashley, Krych, Persel, et al., 1994), as well as with the combined nurse and vocational case manager model (Malec, Buffington, Moessner, et al., 1995).

There were conflicting results about the effects of case management on disability or functional status, living status, family impact, and other aspects, and some findings were mentioned in only one study. The clinical trial resulted in no functional status changes among case managed subjects, despite an extended period of rehabilitation (Greenwood and Brooks, 1994). However, when two forms of case management were compared, both the single and multiple case manager/insurance approaches showed significant functional improvements (Ashley, Krych, Persel, et al., 1994).

Findings also conflicted on the effect of case management on subjects' living arrangements. Greater independence was demonstrated with the single case manager and insurance approach (Ashley, Krych, Persel, et al., 1993), while greater dependence was found with the general case manager model (Greenwood and Brooks, 1994). In the only study that measured the effect of case management on families, the result was that families sought more medical care, and they changed their amount of leisure time (although the direction was not identified) (Greenwood and Brooks, 1994). A modest majority of respondents in the Malec, Buffington, Moessner, et al. (1995) study found the case manager helpful, but the report did not mention whether this rating referred to the nurse case manager and/or the vocational case manager. Single-study findings included lower rehabilitation costs and higher disability payments for the single case manager model (Ashley, Krych, Persel, et al., 1993) and identification of unmet case management needs among adolescents, seniors, and alcoholic clients (Malec, Buffington, Moessner, et al., 1995).

Conclusions for Question 5

From our review we conclude that there is no clear evidence that case management is effective with survivors of TBI and their families, but neither is there clear evidence that it is ineffective. Further research is warranted to resolve this question. It is not possible to directly compare the three studies reviewed because there were almost no similarities in design, sampling, or outcome measures that would provide a basis for comparison. Although all 312 subjects had been diagnosed with brain injury and had evidence of impairment, the samples represent different subpopulations based on injury severity level: moderate, severe, and mixed. For the latter sample, results were not reported by level of injury. Similarly, although there was a controlled intervention of case management provided in each study, there were key differences in the definitions and case management model characteristics. In addition, there were few, if any, specifications about the case managers' training, discipline, experience, and roles. Only one report (Greenwood and Brooks, 1994) mentioned the number of case managers who had provided the intervention (three), and it also was the only report that mentioned the number of clients that were managed by each case manager at one time (20). In no case was there any evidence of reliability or validity testing of the case management approach.

In addition, there are other weaknesses that contribute to the inability to draw conclusions from this small group of studies. One potential source of bias is the lack of control for cointerventions that may have provided service referrals, care continuity, and client and/or family support that simulated case management. This might have occurred, for example, when a study case manager referred the client to another program or service, and it also may have occurred within the family support system. Without controlling for such an effect, it is not possible to attribute any results to formal case management. Another problem is that the studies each provided case management for different amounts of time and at different stages of recovery. Moreover, they utilized different periods of time to measure the outcomes, and the measures did not continue longitudinally for all subjects. This had the statistical effect of reducing the number of subjects who were likely to benefit from the intervention. That is, even if the intervention did have a positive effect, the difference may not have been apparent. Also, because of the team-oriented nature of case management coordination, none of the researchers were able to arrange blind assessments of the subjects. For this reason, it is not possible to know whether there are provider biases associated with care provided for particular subjects or subject groups.

Finally, nothing is known about the quality of the rehabilitation programs associated with the case management models demonstrated in the reviewed studies. Greenwood and Brooks (1994) point out that since the experimental subjects did not progress despite greater rehabilitation service contact time, the cause may have been that the case manager did not have the authority to improve the quality of rehabilitation. Since client and family outcomes may be related only to rehabilitation program benefit, it would be useful to know how to control for rehabilitation program quality to identify confounding factors. Another consideration for possible effect on the targeted outcomes is the ability to access the rehabilitation programs. Since rehabilitation access depends on program availability in the community, economic feasibility for the patient, and the knowledge of others--such as family physicians and emergency care teams--it would be useful to have measures of those environmental conditions as additional outcome analysis controls.

Despite these methodological weaknesses and the incompatible findings, however, there are some observations that can be made from the collective findings of these three studies. First, two of the three studies found significant improvements associated with case management in at least one type of functional outcome (Ashley, Krych, Persel, et al., 1994; Malec, Buffington, Moessner, et al., 1995). This suggests that perhaps the model of case management that was employed in the Greenwood and Brooks (1995) study was simply the wrong model. Second, in the Greenwood and Brooks (1995) study, the dropout rate among subjects without a case manager was higher, which suggests that subjects may have found the service useful in a subjective way. These glimmers of evidence from three controlled studies provide substantive implications for continued research that can improve upon the methods described above.

Future Research on Question 5

Research studies in the future need to test for possible effects of case management that have not yet been identified. We believe it is important to conduct clinical trials that specify and test the extent of contact with the client and family, role training and competence, service-approval authority, screening/rescreening frequency, and influence within the rehabilitation community network. Reliability and validity testing also are recommended for measuring case management. In addition, controls should be in place for isolating possible cointerventions that simulate care coordination. The control variables should include postinjury rehabilitation elements such as settings, types of therapies, amount of contact times, goal achievement records, and other aspects that may directly affect client outcomes.

We suggest that there be improved outcome measures used in case management clinical trial studies for TBI subjects. In addition to outcomes of changed client functionality, there should be outcomes of changed family functionality. Since much of case management communication is directed toward helping family members learn what to expect and where to obtain services, relevant outcomes would include family use of community and rehabilitation services and indicators of family assertiveness regarding care expectations. While case management may only indirectly affect a client's functional outcomes such as level of disability, vocational status, and living status, it is possible that case management can directly affect family knowledge of TBI rehabilitation needs and services, level of psychosocial anxiety, and family competency in coping with TBI.

We also recommend separate measures and analyses for subjects with mild, moderate, and severe disability. Greenwood and Brooks (1994) interpreted their findings that more case-managed-group relatives reported a major TBI effect on the family and had more use of prescribed drugs and medical services by attributing the differences to a more severely ill sample in the case-managed group. However, this could not be verified because they had not controlled for severity of illness. Third, if family members were measured at pre- and postintervention points, the case management intervention effects should become more apparent. Finally, for purposes of study comparability, outcomes could be measured at 12-month postinjury intervals.