A Natural Setting Behavior Management Program for Persons With Acquired Brain Injury: A Randomized Controlled Trial

Experimental Design 

This clinical trial employed a repeated-measures, multiple baseline randomized design with 3 groups. Multiple pretests were employed to provide a more stable assessment of behaviors before and after treatment. The full NSBM group, which received both education and individualized assistance with behavior management, was compared with an education-only and follow-along only control group. Observations and service provision were done according to a weekly schedule displayed in figure 1.

View Large Image Download to PowerPoint

Fig 1. Flow of participants through NSBM trial. NOTE. Three cases were later deleted from analyses (2, goal of increase in behavior frequency; 1, zero frequency of baseline problem behavior).

Participants 

Eligible participants: (1) had a diagnosis of TBI or other acquired brain injury; (2) were at least 16 years of age; (3) had a presenting behavioral impairment causally related to brain injury; (4) were medically stable; (5) did not have a primary psychiatric, developmental, or substance-abuse diagnosis; (6) had an intact and identifiable caregiver system; and (7) were at least 6 months postinjury. The program was grant-funded, so there was no cost to participants.

We decided to include both TBI and other acquired brain injuries for several reasons. People with both TBI and other forms of acquired brain injury exhibit maladaptive or disruptive behavior, and effective interventions need to be developed for both. Our intervention is at the behavioral level and is not in itself a biologic intervention. People with TBI are diverse rather than homogeneous at the level of brain impairment. Many studies in brain injury rehabilitation have involved participants with both TBI and other forms of acquired brain injury.

The flow of participants from recruitment to study completion is described in figure 1 in accordance with CONSORT standards for clinical trials.16 We recruited participants from community-living people with TBI in northern New Jersey by mail, public service announcements, and from lists of discharged outpatients. Over a 3.5-year period, 181 telephone inquiries were received. Potential participants were initially screened by a brief telephone interview. After this screening, 47 cases who met criteria and who were willing to participate were scheduled for an intake assessment in their community (usually, home) setting. Six of these dropped out during initial assessment due to unstable environments, scheduling difficulties, or other listed problems. Forty-one then began the experiment. Only l person subsequently dropped out (going on extended vacation and so becoming unavailable for follow-up behavioral observation). Analyzable data are not available on these dropouts. Forty participants completed the study. In addition, 3 cases were excluded from core analyses for statistical reasons (see Data Analysis section). Subjects were enrolled as they were recruited over the course of the study period.

Demographic and injury characteristics for each of the experimental groups as well as the sample as a whole are presented in table 1. Participants were young to middle-aged adults (average age, 40.5y). The sample was predominantly male and white. Mean time postinjury was 7.6 years, without significant difference across groups. Most (91%) participants reported length of unconsciousness greater than 24 hours, again without differences across groups. Twenty-four participants in the final sample had TBIs (14 involved in motor vehicle collisions, 3 pedestrians hit by a motor vehicle, 3 falls, 2 bicyclists hit by a motor vehicle, 1 construction-related accident, 1 assault); 13 had non-TBIs (5 with anoxia, 3 strokes, 2 arteriovenous malformations, 1 aneurysm, 1 encephalopathy, 1 electrocution). The majority (75.7%) had received postacute rehabilitation.

Table 1. Description of Sample

	Descriptors	Group			Total	Chi-Square or F Test	P*
		Control	Education	NSBM
	n	12	13	12	37
	Demographic descriptors
	Current age (y)
	Mean ± SD	38.6±11.1	42.8±14.6	40.2±11.3	40.5±12.2	0.34	NS
	Sex, n (%)
	Female	2(16.7)	4(30.8)	3(25.0)	9(24.3)	0.74	NS
	Male	10(83.3)	9(69.2)	9(75.0)	28(75.7)
	Race, n (%)
	White	7(58.3)	13(100.0)	11(91.7)	31(83.8)	13.87	NS
	African American	0(0.0)	0(0.0)	1(8.3)	1(2.7)
	Hispanic	3(25.0)	0(0.0)	0(0.0)	3(8.1)
	Asian	1(8.3)	0(0.0)	0(0.0)	1(2.7)
	Other	1(8.3)	0(0.0)	0(0.0)	1(2.7)
	Marital status: current, n(%)
	Single	6(50.0)	2(15.4)	4(33.3)	12(32.4)	7.07	NS
	Married	6(50.0)	7(53.8)	6(50.0)	19(51.4)
	Divorced	0(0.0)	3(23.1)	2(16.7)	5(13.5)
	Widowed	0(0.0)	1(7.7)	0(0.0)	1(2.7)
	Education status: preinjury, n(%)
	9−11	1(8.3)	4(31.8)	2(16.7)	7(18.9)	7.41	NS
	12/GED	2(16.7)	4(31.8)	2(16.7)	8(21.6)
	Trade	1(8.3)	2(15.4)	2(16.7)	5(13.5)
	Some college	3(25.0)	1(7.7)	2(16.7)	6(16.2)
	Associate’s degree	2(16.7)	0(0.0)	1(8.3)	3(8.1)
	Bachelor’s degree	1(8.3)	0(0.0)	1(8.3)	2(5.4)
	Master’s degree	2(16.7)	2(15.4)	2(16.7)	6(16.2)
	Employment status: preinjury, n(%)
	Full-time student	3(25.0)	3(30.8)	4(33.3)	10(27.0)	3.04	NS
	Competitively employed	7(58.3)	7(53.8)	8(66.7)	22(59.5)
	Unemployed	2(16.7)	3(7.7)	0(0.0)	5(13.5)
	Injury descriptors
	Time postinjury (y)
	Mean ± SD	7.3±6.9	7.8±9.4	7.6±6.8	7.6±7.6	0.01	NS
	Etiology, n (%)
	TBI	8(66.7)	12(92.3)	4(33.3)	24(65.9)	10.2	.01
	Other acquired brain injury	4(33.3)	1(7.7)	8(66.7)	13(35.1)
	Duration of unconsciousness, n (%)
	<24h	1(8.3)	0(0.0)	1(8.3)	2(5.4)	3.09	NS
	>24h	11(91.7)	12(92.3)	11(91.7)	34(91.9)
	Never unconscious	0(0.0)	1(7.7)	0(0.0)	1(2.7)
	Postacute rehabilitation, n (%)
	Yes	11(91.7)	7(53.8)	10(83.3)	28(75.7)

Abbreviations: GED, General Educational Development diploma; NS, not significant (P>.05); SD, standard deviation.

Using chi-square tests (for categorical descriptors) and F tests (for interval-level descriptors) we found no significant differences in demographic or injury descriptors across groups—with 1 possible exception. The proportion of TBIs in the education-only group appears to be different from that in the overall group, at a nominal P less than .01 level. In interpreting this figure, it is important to address the fact that a total of 11 such post hoc comparison tests were run. (In addition to the 10 tests in table 1, we tested whether TBIs and acquired brain injuries differed in frequency of target behaviors at baseline and found that differences were not statistically significant.) If one controls for experiment-wide error at the .05 level in the testing of multiple differences between the groups at baseline, single comparisons must reach a significance level of .0046 to be declared significant using the Bonferroni criterion. In other words, given the number of post hoc tests done, the nominally significant bivariate difference may have been due to chance. We strictly followed a blind randomization procedure, so the difference in proportions is attributable to chance. Difference in improvement between TBIs and acquired brain injuries did not attain statistical significance, and post hoc covariance analyses also showed that results were insensitive to the traumatic versus acquired brain injury distinction.

Intervention 

The NSBM team was composed of 2 doctoral-level clinical psychologists and a master’s-level behavior technician. An experienced clinical neuropsychologist (principal investigator) supervised all team members and reviewed all assessments and treatment protocols.

The intervention began with a 3-week assessment phase during which we identified target problem behaviors for all participants and family members. Thereafter, education was provided to all but control participants. For full NSBM condition, individualized treatment planning was also provided.

Data Collection Procedures 

All assessment, data collection, education, and intervention sessions were approximately 2 hours long. NSBM staff met with the participants and caregivers in their natural setting (eg, home, school, or day treatment facility). In the majority of cases, the source of historical and medical information was confirmed by a review of the participant’s medical records. Duration of unconsciousness was coded to be consistent with the national Traumatic Brain Injury Model System (TBIMS) dataset (ie, either <24h or >24h).

Measures 

We collected background demographic data using surveys and procedures of the TBIMS dataset.18 Participants and their caregivers were the primary sources of information, and in most cases the participant’s medical records were available to verify data. The behavioral target data measured were observable aggressive behavior or other forms of disinhibited behavior. NSBM staff worked with participating families to develop chart assessment procedures to monitor the frequency of target behaviors.

Data on caregiver burden and stress were also gathered at each major observational point. Subscales from the Questionnaire on Resources and Stress for Families with Chronically Ill or Handicapped Members (QRS) was employed to assess stress of family participants. Specific subscales used were 3 (limits on family opportunities, short form), 7 (pessimism, long form), and 11 (personal burden for respondent, short form).19 An adapted version of the Maslach Burnout Inventory (MBI) was used to assess caregivers’ burden or burnout.20 The MBI is the most widely used scale of the experience of burnout and has 3 subdimensions: emotional exhaustion, depersonalization, and dissatisfaction over reduced personal accomplishment.

The Neurobehavioral Functioning Inventory–Revised (NFI-R) is a 70-item scale designed to assess behavioral dysfunction after brain injury.21 Two versions are administered to gather problem frequency information about the person with brain injury: a participant version and a family and caregiver version. Items are rated on a 5-point scale (1, never; 2, rarely; 3, sometimes; 4, often; 5, always). The NFI has 6 subscales, based on factor analytic methodologies: depression, somatic complaints, memory/attention difficulties, communication deficits, aggressive behaviors, and motor impairment.

Data Analysis 

Analysis of group behavioral outcomes data confronts certain difficulties. Maladaptive behavior can be highly variable in its frequency, giving rise to skewed, irregular, or heteroscedastic distributions. The targeted behavior problems were defined in individualized terms, so a reduction in frequency of 1 means different things for different participants.

To deal with these issues, we first averaged the 3 pretest values to obtain a more stable baseline value, a recommended procedure for analysis of repeated-measures data.22 Frequencies of behavioral outcomes were then transformed into percentage of average baseline frequency.

Although analysis of covariance (ANCOVA) is commonly recommended for analysis of change data, its application to the behavioral outcomes data in the current study is questionable for reasons given above.22, 23 We chose to employ 2 simpler methods used in previous research on behavior counts. The first method involved simply classifying cases as improved or not from baseline. This rating was quantitatively derived from frequency counts and did not involve subjective analysis. A chi-square exact statistic was then employed to test whether improvement rates differed significantly across groups.a The second analysis method employed percentage change in problem behavior frequency as the outcome variable. One-way analyses of variance (ANOVAs) were run at each time period to evaluate the significance of between-group differences.

Three cases were deleted from most analyses. Two participants with behavioral deficits had goals of increasing behavior frequency: these cases can be compared with others in terms of improvement or not but not in terms of diminished frequency of behavior challenges. One case was also dropped because a baseline count of observed target behavior was too low (zero) for computation of percent change.

We employed ANCOVA for analysis of between-group differences in stress and caregiver burden measures at times 4, 5, and 6. ANCOVA is justifiable for these variables because the same substantive outcome scale is assessed across participants, and pretreatment levels of burden, stress, and neurobehavioral functioning may be expected to affect post-treatment levels. Statistical programs employed were SPSS, versions 10 and 12.a

Changes in Frequency of Behaviors 

Effects of treatment on frequency of target problem behaviors were analyzed in several ways. First, targeted behaviors improved more frequently among NSBM participants than among education or control participants at all 3 follow-up points, as shown in table 2. These differences were not statistically significant at the point at which education ended (7wk post baseline) or at termination of individualized behavioral management (16wk post baseline), but larger differences appeared at the 3-month follow-up point (30wk post baseline), and these differences were statistically significant (P<.002).

Table 2. Improvement in Target Behavior at Times 4, 5, and 6

	Group	Percentage (n/N) of Participants Improving
		7 Weeks Post Baseline	16 Weeks Post Baseline	30 Weeks Post Baseline
	Control	54.5(6/11)	41.7(5/12)	66.7(8/12)
	Education	53.8(7/13)	25.0(3/12)	40.0(4/10)
	NSBM	70.0(7/10)	50.0(5/10)	100.0(10/10)
	χ22 exact test (P value)	.73(<.075)	1.54(<.56)	8.42(<.002)

Changes in frequency of targeted behaviors from baseline were analyzed by ANOVA. Differences between the 3 groups were not significant at the first 2 follow-up points but were significant (F=3.32, P=.05) at the last follow up point (30wk post baseline). Post hoc analysis revealed a significant difference between the NSBM and the education-only groups (Tukey honestly significant difference, P<.04) but other differences did not reach statistical significance. Figure 2 portrays these changes in frequency of target behaviors in terms of percentage change from baseline rates. The same pattern is evident as was reported for change-or-not: differences between groups were greatest at the last post-treatment point. It is worth noting that baseline observations strongly predicted outcomes: average frequency of targeted behavior at baseline correlated with frequency at 7, 16, and 30 weeks post baseline (r=.81, r=.76, r=.75; all P<.001).

View Large Image Download to PowerPoint

Fig 2. Percentage change in frequency of target behaviors from baseline for 3 groups over time.

Secondary Outcomes 

Changes in other outcomes were examined post hoc using ANCOVA at each point in time controlling for baseline values as a covariate. While no significant differences between groups were found for QRS subscales at any of the time points, potentially significant differences were found for MBI and NFI subscales.

The emotional exhaustion subscale of the MBI was examined first. As one would expect, baseline ratings of emotional exhaustion strongly affected emotional exhaustion outcome scores (F=150.0, F=70.9, F=59.5; all P<.000 for the 3 outcome points, respectively). Controlling for baseline emotional exhaustion, treatment effects failed to reach significance at 7 and 16 weeks post baseline (F=3.05, P<.07; F=1.28, P<.29, respectively) but did reach significance at the last follow-up point (F = 4.26, P<.03). Post hoc 2-group ANCOVAs at this point revealed a significant difference (F=10.54, P<.005) in emotional exhaustion between control and education only cases (mean: 22.0 vs 32.2, respectively); emotional exhaustion in NSBM did not differ significantly from other groups (mean, 28.5). MBI depersonalization and personal accomplishment at intake correlated highly with subsequent ratings of the same scales, but treatment effects were not significant.

While initial ratings of NFI aggression as rated by caregivers strongly correlated with aggression outcomes (r=.67, r=.66, r=.85; all P<.001, at 7, 16, and 30wk post baseline), no significant effects of treatment on these ratings of aggression were found. When ratings by the person with acquired brain injury were examined, initial ratings of aggression again strongly affected outcomes (r=.79, r=.75, r=.85; all P<.001), but again treatment group made little difference. Although ratings on the aggression subscale of the NFI consistently declined over time in the NSBM group, the trend did not reach statistical significance at any of the outcome points. No significant effects on other NFI subscales were detected.

Study Limitations 

A major limitation of the current study was the small sample size. With a larger sample, the unequal distribution of TBIs versus acquired brain injuries across treatment conditions would in all probability have been remedied. Although data in the current study do not indicate that this irregularity explains results, one can still speculate that our results were affected by uncontrolled interaction effects between etiology and treatment variables. With a larger sample, one might employ a stratified randomization procedure. For instance, if one believed that etiology is an outstanding factor affecting response to behavioral interventions, one could separately randomize acquired brain injuries versus TBIs. The solution is a larger future study, which should include both TBIs and acquired brain injuries.

The volunteer participants in the current study may also differ from other clinical populations with TBI and acquired brain injury; whether differences in wealth or insurance coverage affect the effectiveness of NSBM is not known. The sample studied had high base rates of aggressive behavior and caregiver distress, and caution should be raised in generalizing findings to less behaviorally challenged persons and their caregivers.

Finally, group studies of behavioral interventions involve analysis of data on heterogeneous behaviors as well as individualized treatment approaches. Group studies, especially RCTs, are difficult to conduct because of these problems. Any focus on average outcomes masks variations in the nature and extent of individual responses to treatment. As a consequence—and perhaps also because of the paucity of research funding for randomized trials—single case designs have most commonly been employed in behavioral research.26 Nonetheless, we maintain that well-controlled group studies are needed to address the question of whether behavioral interventions are helpful to persons with TBI and their caregivers as a group. RCTs provide a recognized standard of evidence which individual case studies cannot offer27, 28, 29 and are needed to complement—but not replace—the knowledge provided by single subject research designs.