Team Training and Stroke Rehabilitation Outcomes: A Cluster Randomized Trial

Participants 

All VA hospitals with active inpatient (acute or subacute) rehabilitation units that treated a minimum of 2 stroke patients per month and that subscribe to the VA Functional Status Outcomes Database (FSOD)19 were eligible for and recruited for inclusion in the trial. The FSOD uses the FIM instrument and classifies function on a scale of 1 (total dependence) to 7 (independence) on each of 18 items; 13 of these items are related to motor function, and 5 are related to cognitive function.20, 21 The FIM has been shown to show good reliability and validity21 and it has been used extensively22, 23 to assess patient outcomes. All teams had representatives from 6 professional disciplines—medicine, nursing, occupational therapy (OT), speech-language pathology (SLP), physical therapy (PT), and case management/social work. Patients were included if their demographic and assessment data were reported to the FSOD by one of the participating units, with discharge date within either the baseline or follow-up assessment period for that site. Patients were excluded if they had invalid or missing FIM assessments; fewer than 2 FIM assessments; or LOS of 3 days or less. Primary analyses were limited to patients with a diagnosis of stroke, and time since onset of 90 days or less. Additional analyses were conducted on all patient data, regardless of primary diagnosis. Each study site obtained approval from the local VA research committees and affiliated institutional review boards (IRBs).

Interventions 

The 6-month team training intervention occurred in 3 phases and involved 227 rehabilitation staff. It began with a concentrated 2.5-day team training workshop designed for 2 self-identified team leaders from each of the 15 experimental sites. Rehabilitation medicine (13 physicians or osteopaths) was the most commonly represented discipline among the team leaders. Other participating leaders included nursing (2 registered nurses), PT (4 therapists), OT (2 therapists), kinesiotherapy (3 kinesiologists), social work (1 social worker/case manager), SLP (3 speech-language pathologists), and 2 administrators. The workshop emphasized skill development in team problem-solving strategies and the use of program evaluation data. The second phase occurred 3 to 5 weeks after the workshop and consisted of written action plans to address team process problems based on discussions at the earlier workshop. In the third phase (months 3 to 6), workshop participants received telephone and videoconference consultation (eg, advice on implementation of action plans, facilitation of team process skills).

Treatment implementation, a research methodology commonly used in psychosocial interventions, guided the design and implementation of our team training intervention. Using the framework of Lichstein et al,24 3 categories of treatment implementation were used and monitored: delivery, receipt, and enactment. Treatment implementation methods helped to ensure consistent and reliable delivery of the intervention by research staff and to document skill use by participants. A complete description of the treatment implementation methods used in this study along with specific content and evidence for the integrity and fidelity of the team training intervention is presented elsewhere.25

Individual team leaders in both the team training and information only groups received summaries of their own team’s performance on process measures taken preintervention and again at the end of the intervention. These summaries included comparison data collected from the larger sample and again after the completion of the intervention. The summary documents also contained suggestions on how to use these data and options for individual consultations with project staff. Financial resources ($1000 per site) were provided to offset expenses incurred during the study. Suggestions were offered on how the team process reports and funds could be used to promote effective team functioning. These activities defined the standard feedback condition.

The inpatient services use a multidisciplinary team approach to evaluate and treat rehabilitation patients. This is the standard of care and consistent with guidelines of the Center for Medicare & Medicaid Services of the U.S. Department of Health and Human Services1 and CARF.2 Core disciplines represented on the team generally include medicine, nursing, PT, OT, SLP, and social work or case management.

Outcomes 

The primary outcome of this trial was the change from admission to discharge in the motor skills component of the FIM score. Secondary outcomes were LOS (measured in days) and discharge disposition (discharged to home or discharged to other location).

Outcome data on discharged patients were collected from the VA FSOD. In addition, we collected patient demographic (age, sex, race and ethnicity, marital status), clinical (diagnostic class), and treatment (admission class, program interruption) characteristics. FIM assessments were done by clinical staff at each facility in the standard manner and consistent with published guidelines of the Unified Data Systems.26 Staff from each site performed FIM assessments and participated in the VA FSOD for a minimum of 5 years prior to this study.

For each participating site, we collected data on all patients admitted during the 12-month period prior to the intervention (preintervention) and during the 12-month period after the completion of the intervention (postintervention). Because of the nature of the intervention, these periods were staggered across participating sites.

Case-Mix Severity 

To account for differences in admitting motor FIM score across study groups, all comparisons of outcomes were adjusted for functional-related groups (FRG) category. FRGs account for case mix severity27, 28 by assigning patients to 1 of 9 categories according to admission motor FIM score and age; the first (lowest severity) category was used as a reference category in models.

Randomization 

We stratified the 31 participating sites into 4 groups according to volume of patients (above or below median) and average admission FIM of stroke patients (above or below median); volume and FIM scores were based on stroke patients discharged for the year leading up to September 2000. Within each stratum we randomized sites to either intervention or control group using a computer; each stratum was force randomized to have 4 sites in 1 arm.

Sample Size 

Because the patients were clustered within study sites, we calculated our statistical power using methods appropriate for cluster-randomized outcomes.29 Using data from 45 VA rehabilitation sites which participated in a previous observational study18 we estimated the intracluster correlation in motor FIM score gain to be 4.6% and the standard deviation (SD) to be 12.8; assuming at least 20 patients per site, we anticipated that with 28 sites we would have 80% power to detect a difference in motor FIM gain between study arms of 6 points.

Statistical Analysis 

We used analytic techniques appropriate for cluster-randomized trials.30, 31 First, we summarized patient characteristics by study arm and overall, and tested for differences between the 2 arms using cluster-adjusted chi-square and t tests.30

When we assessed change in motor FIM scores for normality we identified anomalies indicating that these scores were potentially biased upward for patients with negative gain. That is, the number of patients with gain exactly equal to zero was much higher, and the number of patients with negative gain much lower, than expected by chance, and moreover, the rate of zero gains was twice as high in the control group as in the intervention group (P<.001). This is the pattern one would expect if patients with negative gain (ie, discharge FIM score lower than admission FIM score) were being reported as zero gain. Thus, to avoid potential bias toward the null caused by upward rounding of the outcome, we dichotomized motor FIM gain for each patient as either above or below the overall median of 23 points gained. This dichotomized outcome was used for all analyses of the primary outcome.

To assess the effect of the intervention on the 3 patient outcomes we used hierarchical generalized linear models. Such models are appropriate when comparing outcome measures that have been randomized in clusters,30, 31 and allowed us to adjust for patient characteristics that were imbalanced across study arms. For each outcome we estimated either a logistic (motor FIM gain above median and discharge disposition) or linear (LOS, log transformed to reduce skewness) hierarchical generalized linear model, with each outcome measure assumed to vary both within and across study sites. Each model included an indicator variable for study condition (intervention vs control), and indicator variable for measurement period (preintervention or postintervention), and an indicator representing the interaction between these 2. By testing the hypothesis that the interaction term was different from zero, we were able to test whether the intervention had an effect on team performance as measured by each patient outcome. For example, to test the hypothesis that the intervention increased the probability of patients gaining more than 23 points, we estimated the following model: Let Yij indicate wither the ith patient at the jth site had a gain of at least 23 points on motor FIM; let P indicate whether that patient was admitted pre- (P=0) or post- (P=1) intervention, and let E indicate whether that patient was treated at a control (E=0) or intervention (E=1) study site. Then we estimated:

The primary analysis included only stroke patients with time since onset of 90 days or less. We replicated the analysis using all patients discharged during the data collection periods.