Archives of Physical Medicine and Rehabilitation
Volume 89, Issue 5 , Pages 974-981, May 2008

Prediction of Memory Rehabilitation Outcomes in Traumatic Brain Injury by Using Functional Magnetic Resonance Imaging

  • Gary E. Strangman, PhD

      Affiliations

    • Department of Psychiatry, Massachusetts General Hospital, Harvard Medical School, Charlestown, MA
    • Department of Physical Medicine and Rehabilitation, Spaulding Rehabilitation Hospital, Harvard Medical School, Boston, MA
    • Corresponding Author InformationReprint requests to Gary E. Strangman, PhD, Neural Systems Group, 149 13th St, Psychiatry, Room 2651, Charlestown, MA 02129
    • ,
  • Therese M. O'Neil-Pirozzi, ScD

      Affiliations

    • Department of Physical Medicine and Rehabilitation, Spaulding Rehabilitation Hospital, Harvard Medical School, Boston, MA
    • Department of Speech-Language Pathology and Audiology, Northeastern University, Boston, MA
    • ,
  • Richard Goldstein, PhD

      Affiliations

    • Department of Physical Medicine and Rehabilitation, Spaulding Rehabilitation Hospital, Harvard Medical School, Boston, MA
    • ,
  • Kalika Kelkar, BA

      Affiliations

    • Department of Physical Medicine and Rehabilitation, Spaulding Rehabilitation Hospital, Harvard Medical School, Boston, MA
    • ,
  • Douglas I. Katz, MD

      Affiliations

    • Department of Neurology, Boston University School of Medicine, Boston, MA
    • Braintree Rehabilitation Hospital, Braintree, MA
    • ,
  • David Burke, MD

      Affiliations

    • Department of Rehabilitation Medicine, Emory University, Atlanta, GA
    • ,
  • Scott L. Rauch, MD

      Affiliations

    • McLean Hospital, Harvard Medical School, Belmont, MA
    • ,
  • Cary R. Savage, PhD

      Affiliations

    • Hoglund Brain Imaging Center, Kansas University Medical Center, Kansas City, KS.
    • ,
  • Mel B. Glenn, MD

      Affiliations

    • Department of Physical Medicine and Rehabilitation, Spaulding Rehabilitation Hospital, Harvard Medical School, Boston, MA

Article Outline

Abstract 

Strangman GE, O'Neil-Pirozzi TM, Goldstein R, Kelkar K, Katz DI, Burke D, Rauch SL, Savage CR, Glenn MB. Prediction of memory rehabilitation outcomes in traumatic brain injury by using functional magnetic resonance imaging.

Objective

To evaluate the ability of functional magnetic resonance imaging (fMRI) measures collected from people with traumatic brain injury (TBI) to provide predictive value for rehabilitation outcomes over and above standard predictors.

Design

Prospective study.

Setting

Academic medical center.

Participants

Persons (N=54) with TBI greater than 1 year postinjury.

Intervention

A novel 12-session group rehabilitation program focusing on internal strategies to improve memory.

Main Outcome Measure

The Hopkins Verbal Learning Test−Revised (HVLT-R) delayed recall score.

Results

fMRI measures were collected while participants performed a strategically directed word memorization task. Prediction models were multiple linear regressions with the following primary predictors of outcome: age, education, injury severity, preintervention HVLT-R, and task-related fMRI activation of the left dorsolateral and left ventrolateral prefrontal cortex (VLPFC). Baseline HVLT-R was a significant predictor of outcome (P=.007), as was injury severity (for severe vs mild, P=.049). We also found a significant quadratic (inverted-U) effect of fMRI in the VLPFC (P=.007).

Conclusions

This study supports previous evidence that left prefrontal activity is related to strategic verbal learning, and the magnitude of this activation predicted success in response to cognitive memory rehabilitation strategies. Extreme under- or overactivation of VLPFC was associated with less successful learning after rehabilitation. Further study is necessary to clarify this relationship and to expand and optimize the possible uses of functional imaging to guide rehabilitation therapies.

Key Words: Magnetic resonance imaging, Rehabilitation

 

COGNITION IS FREQUENTLY disrupted after traumatic brain injury (TBI), and many resulting impairments persist over time and contribute to psychosocial problems. Although the effects of TBI can be variable, the domains of memory,1, 2, 3 attention,4 and executive functioning5, 6, 7, 8, 9 are quite consistently impaired. Persisting memory dysfunction is among the primary sources of disability after TBI and is a major factor in failure to return to work,3 a predictor of unsuccessful completion of rehabilitation and vocational training,10 and a serious obstacle to improving quality of life.11, 12

Stuss and Gow13 have noted that the patterns of cognitive impairment observed in TBI are similar to those found in nontraumatic focal injury to the prefrontal cortex. Memory studies14, 15, 16, 17 in subjects with frontal lesions point in particular to abnormalities in the strategic aspects of memory, which are closely tied to executive functioning and to prefrontal cortex. For example, subjects with frontal lesions have difficulty using semantic encoding strategies during verbal learning.16 Cognitive activation neuroimaging studies after TBI have consistently supported the role of prefrontal system dysfunction in TBI.18 This includes frontal alterations after TBI in a continuous performance task,19 Wisconsin Card Sorting Test,20, 21 and several studies with working and verbal memory tasks.22, 23, 24, 25

Four prospective, randomized controlled studies26, 27, 28, 29 of the effectiveness of memory rehabilitation performed by using internal (and, in 2 cases, also external) strategies for subjects with TBI have been identified. These studies provide evidence that training people with TBI on internally based memory strategies can significantly improve memory function. However, substantial difficulty arises when considering group analyses on the efficacy of interventions in people with TBI; the population of people with TBI is heterogeneous with regard to severity and localization of cerebral damage and the respective consequences of the injury30, 31 as well as psychologic and socioeconomic background.32 As a result, one cannot expect that patients should all be treated with a similar approach. Indeed, several studies provide evidence of differential treatment outcomes based on individualized programs or client characteristics,26, 27, 28, 33 including, in particular, the severity of impairment.26, 27, 28, 34, 35

An important avenue of further study is whether variability in rehabilitation outcomes can be reduced by using approaches that can take into account the specific cognitive, neuroanatomic, and neurophysiologic consequences of TBI on each survivor. Cognitive consequences can be assessed through neuropsychologic testing. Functional neuroimaging (FNI) introduces a separate modality for adding individualized elements to cognitive assessment and treatment, the ability to observe the specific, dynamic cerebral activation patterns during cognitive challenge. We posited that when faced with a mental challenge similar to that delivered by a given rehabilitation program, poorly performing patients may nevertheless reveal brain activity patterns consistent with good outcome, whereas otherwise similar patients may reveal activity patterns consistent with poor outcome. We hypothesized, therefore, that predicting rehabilitation outcomes might be aided by identifying brain activations predictive of outcome.

A recent longitudinal case series by Laatsch et al36 who found that functional magnetic resonance imaging (fMRI) activation during probe tasks pre- and postcognitive behavior therapy was variable across people with TBI. This report highlighted the variability in response to therapy and the sensitivity of FNI for detecting individual differences, an important ingredient if it is to be used for individualizing neurorehabilitation. Additional support for our hypothesis comes from Musso et al37 who found a significant relationship between comprehension improvement with strategic language training and blood flow in both the right superior temporal gyrus and the left precuneus cortex for patients with Wernicke's aphasia. This more directly suggested that FNI might be useful specifically for predicting rehabilitation outcomes.

Studies seeking to predict outcomes by using FNI predictors still remain uncommon. However, Small et al38 showed that neural activation of the cerebellum contralateral to a unilateral stroke was an even better predictor of motor outcome than was the classic approach to stroke prognositication, the evaluation of motor function on the impaired side. In a study of patients with a first lacunar infarct of the pyramidal tract, Loubinoux et al39 found that activation in the ipsilesional M1, S1, and insula during a flexion and extension movement of the impaired hand was associated with better hand function 1 year after the stroke. Hoeft et al40 recently showed added predictive value of FNI measures, over and above standard behavioral measures, for predicting success in learning to read in children identified at high risk for reading impairment. These initial findings all suggest a role for FNI in helping to predict individual learning outcomes.

Previously, we developed a paradigm to examine brain regions supporting the use of strategy-based verbal learning.41 We measured blood flow with positron emission tomography (PET) while healthy participants learned related and unrelated word lists. The related word lists could be mentally reorganized into 4 semantic categories during learning. There were 3 primary scanning conditions: (1) spontaneous: participants heard a list of related words but were not given a learning strategy, (2) directed: participants heard a word list and were explicitly instructed to notice semantic relationships and mentally group related words together to improve memory (a semantic clustering strategy), and (3) unrelated: participants heard a list of unrelated words. In graded linear PET contrasts, 2 activations were found in the left prefrontal cortex. Findings indicated that regions in the left prefrontal cortex showed progressively increased blood flow as participants received increasing guidance regarding the correct strategy. An fMRI study by Logan et al42 expanded this approach to study changes in memory in normative aging. They found that, in contrast to younger subjects, elderly participants underrecruited regions in the left prefrontal cortex in a spontaneous encoding condition. However, this failure to activate the left prefrontal cortex was normalized by instructing elderly participants to use a semantic elaboration strategy, suggesting that memory problems with normative aging reflect difficulty spontaneously recruiting available prefrontal resources. Taken together, these 2 functional imaging studies indicate that the use of a semantic organization strategy during learning, and the associated patterns of brain activation, can be encouraged by providing instruction in strategy use before the learning trials.

Based on these studies, we speculated that the extent to which a given type of intervention activates critical brain regions could be used to help determine whether the intervention might be effective for people with TBI in general, for certain subpopulations, or, more importantly, for specific people. Our specific hypothesis was as follows: the ability to activate previously identified key brain regions (particularly the left dorso- and ventrolateral prefrontal cortex) during a directed encoding condition would provide predictive information regarding intervention outcomes. The usefulness of FNI for these purposes, of course, depends on its ability to add predictive value for outcome over and above the information that is available from more readily available predictors. No studies have yet been reported that assess the ability of FNI to predict rehabilitation outcome after TBI. The goal of the present study, therefore, was to evaluate the ability of a specific fMRI paradigm to provide added predictive value for outcomes from a specific rehabilitation program over and above standard predictors of outcome.

Back to Article Outline

Methods 

Participants 

Fifty-eight participants with a TBI were enrolled following a written informed consent procedure approved by the Human Research Committee at the Massachusetts General Hospital. Participants were recruited from mailings to clients of The Commonwealth of Massachusetts Brain Injury and Statewide Specialized Community Services Department, members of the Brain Injury Association of Massachusetts, local support groups, and patients of study-affiliated physicians. All participants were (1) at least 18 years of age at the time of injury, (2) sustained a TBI of any severity at least 12 months before the study, (3) were right-handed and fluent in English, and (4) self-reported as having difficulty with memory after their injury. Exclusion criteria included a score of less than 4 on either the expression or comprehension items of the FIM, nontraumatic etiology of cerebral dysfunction in addition to TBI, active major illnesses, preinjury history of psychiatric disease, inability to read single words at an eighth-grade reading level, and current drug or alcohol dependence (criteria from the Diagnostic and Statistical Manual of Mental Disorders, 4th Edition). We excluded 2 participants who did not complete at least 8 of 12 memory intervention sessions. Because of uncorrectable motion (>3mm) during scanning sessions, 2 additional participants needed to be excluded from the analysis. We proceeded with analysis of 54 participants. Demographic information for this group appears in table 1. Injury severity information appears in table 2.

Table 1. Demographic Characterization of Study Participants
CharacteristicsParticipants With TBI (N=54)
Mean ± SDMinMax
Age (y)47.6±11.124.365.4
Education (y)14.5±2.21020
Time since injury (y)11.7±9.91.337.6
LOC (d)13.3±25.20105

Abbreviations: LOC, loss of consciousness; SD, standard deviation.

Table 2. Severity Assessments for Study Participants
AssessmentsParticipant CountPercentage of Total
Mild1425.9
Moderate1120.4
Severe2444.4
Insufficient data59.3
Total54100.0

Procedures 

The experiment involved 16 sessions. Session 1 (pretest) consisted of a series of standardized evaluations of memory, executive function, and language including the Hopkins Verbal Learning Test−Revised (HVLT-R), Rivermead Behavioural Memory Test−II (RBMT-II), Boston Naming Test 2nd Edition Short Form, the Boston Diagnostic Aphasia Examination, animal naming, the Trail-Making Test parts A (TMT-A) and B (TMT-B), and the Weschler Memory Scale−Revised digit span (forward and backward). We selected HVLT-R forms 1, 5, and 6 and RBMT-II forms A, B, and C for use in this study.

In session 2, approximately 3 days later, each participant underwent fMRI scanning during verbal encoding.41 Six word lists were presented: 1 list was a run in which stimuli alternated between a fixation cross (16s) and a list of 16 nouns, 1 word at a time, 2 seconds a word. The list repeated 3 times, with words always appearing in the same order. Immediately after each scan, participants were asked to (1) freely recall as many words as they could remember and then (2) perform a yes-no recognition test of a 32-item list containing the 16 list words plus 16 previously unseen words.

To influence the extent of strategic (specifically, semantic) processing during encoding, word lists were learned under the 3 conditions described earlier: unrelated, spontaneous, and directed. In the unrelated condition, words were semantically unrelated, and participants were simply asked to remember as many words as they could. In the spontaneous condition, the 16 words were grouped into 4 semantic categories (eg, fruits, tools, musical instruments, gemstones) of 4 words each, and participants were again simply given instructions to try to remember as many words as they could.

In the directed condition, the word lists were similar to those in the spontaneous condition. However, immediately before the directed condition, while still in the scanner, participants were trained to use a semantic clustering strategy. Participants were shown 3 categories to which words in a 12-item training list would belong and were asked to group the words together into these 3 categories as they learned the words. After a single presentation of the training list, participants were asked to recall the words from the list, cued by the category names. All subjects were able to recall at least 2 words from each of 2 categories. The final instructions to the participant were as follows: “The next list will be grouped into categories. They will not be the categories you just practiced. Please try to group the next words together into categories as you learn the words.”

Two runs were administered for each condition. To minimize the number of instruction changes, we administered the 2 runs for each condition sequentially. Order effects were partially addressed by counterbalancing the unrelated condition; half of the subjects participated in the order unrelated/unrelated/spontaneous/spontaneous/directed/directed, whereas the other half participated in the order spontaneous/spontaneous/directed/directed/unrelated/unrelated. We did not counterbalance the spontaneous and directed orders because exposure to the directed condition would contaminate any subsequent attempt at gathering data under a spontaneous condition. In this article, we focus on data from the directed condition only.

Sessions 3 through 14 were memory intervention sessions led by one of the investigators based on internal memory strategy evidence-based approaches identified by Cicerone et al.43 We used a group intervention with 3 to 6 members a group cycle and 3 group facilitators (ratio of facilitator to patient of 1:1 to 1:2). Each group cycle ran twice weekly for 6 weeks, 90 minutes a session, for a total of 12 sessions (18 total contact hours). Sessions emphasized semantic organization and other internally based strategies (eg, elaboration and imagery) from encoding, storage, and retrieval perspectives.

Session 15 (posttest 1) was conducted within 3 days after the last memory intervention session and repeated the memory and executive function tests from session 1. Session 16 (posttest 2) was conducted 1 month after session 15 and again repeated the same cognitive tests. No fMRI scanning was performed at either posttest session. Different forms of the various tests were fully counterbalanced across participants and sessions. Pre- and posttest evaluations were conducted by different experimenters with regular cross-validation of test administrators.

Outcome Measures 

Hopkins Verbal Learning Test−Revised 

In the HVLT-R,44 a person is verbally presented a list of 4 words from each of 3 semantically related groups in a semantically unrelated order. Under various time-dependent conditions, the subject is asked to recall and recognize the words presented. The HVLT-R has been normed on healthy subjects age 16 to 92 years and used for people with TBI.45, 46 We selected the HVLT-R as a primary outcome measure over other memory tests because (1) it emphasizes semantic association, which is a focus of our intervention; (2) it is widely used; (3) it has multiple parallel versions, enabling multiple testing sessions with minimal practice effects47; and (4) it can provide multiple quantitative assessments of memory performance.

Rivermead Behavioural Memory Test−II 

The RBMT-II48 is a widely used test designed to be an ecologically valid, broad measure of impairment in everyday memory functioning.49 Test items include asking a patient to remember a hidden belonging, to remember to ask about an appointment, and to deliver a message. The RBMT-II has been standardized on healthy subjects aged 11 to 94 years and on people with brain injury aged 14 to 69 years. It has 4 parallel versions with good alternate form reliability.50

fMRI Data Acquisition 

fMRI was performed with a Siemens Avanto 1.5T scannera with a 12-channel total imaging matrix head coil. Six gradient echo, echo-planar BOLD scans (repetition time [TR]=2.0s; echo time [TE]=40ms; flip, 15°; axial slices, 23; thickness, 6mm; skip, 1mm; in-plane resolution, 3.125×3.125mm) were performed during word list encoding, 1 scan (85 volumes acquired) a word list. Magnetization-prepared rapid-acquisition gradient echo sequences were collected for high-resolution anatomy (TR=1.91s; TE=4.13ms; inversion time, 1.1s; flip, 15°; resolution, 1×1×1mm), along with fluid-attenuated inversion recovery, susceptibility, and diffusion scans for neuroradiologic evaluation.

General fMRI Data Analyses 

fMRI data were preprocessed as follows: (1) realignment, (2) detrending each fMRI time series (≥96s period),51 (3) spatial smoothing (8-mm fluid width at half maximum Gaussian kernel), (4) contrast estimation (percentage of fMRI signal change for task vs fixation), and (5) normalization to Talairach template. Normalization was achieved through a simple 12-parameter affine transformation for all subjects to minimize potential distortions from injured tissue.52 In addition to the 2 excluded participants, 10 individual runs (out of 348 total fMRI runs) were excluded from the analysis because of motion. Data for these 10 runs were replaced by multiple imputation (see later).

Cognitive Data Analysis and Regression Model Variables 

Our primary outcome was the HVLT-R delayed correct recall score (range, 0–12).53 The secondary outcome was the RBMT-II standardized profile score (range, 0–24). Our main interest clinically was whether we could predict outcome beyond the immediate posttest because these reflect more stable as opposed to transitory changes. Moreover, Berg et al26 showed that participants performed better 1 month after the completion of their rehabilitation program. Hence, posttest 2 was prospectively chosen as the primary time point for outcome measurement.

Four standard predictors of long-term outcome were included in each regression model: age (years at time of scanning),54, 55, 56, 57, 58, 59 preinjury education (in years),56, 58, 59, 60, 61 injury severity,54, 55, 56, 57, 59, 61, 62, 63 and behavioral baseline57 (pretest HVLT score or pretest RBMT score). Injury severity was categorized as mild, moderate, or severe based on all data available in the participant's medical records. Injury severity was predominantly (ie, in 85% of cases) based on the duration of loss of consciousness (LOC),54, 55, 63 where mild was 0 to 30 minutes LOC (unless posttraumatic amnesia exceeded 24 hours), moderate LOC was greater than 30 minutes but less than or equal to 24 hours, and greater than 24 hours LOC was designated as severe. In the absence of LOC data, the determination was made based on Glasgow Coma Scale scores (severe, <9; moderate, 9–12; mild, 13–15).54, 59 For 5 participants, we were unable to obtain sufficient, corroborated information about the injury to assess severity, so these data were imputed along with the 10 excluded runs of fMRI data.

In addition to these standard predictors, we included 1 fMRI-based predictor per model fit. Based on neuroimaging studies of verbal encoding,64 strategic memory,40, 41 cognitive training,65 and TBI,18 we chose the average percent fMRI signal change in the left dorsolateral prefrontal cortex (DLPFC) and the left ventrolateral prefrontal cortex (VLPFC) during the directed condition (relative to fixation baseline) as primary predictors of outcome. DLPFC and VLPFC were defined as spherical regions 12mm in radius centered on Talairach coordinates −43,10,28 and −39,35,10, respectively, based on the peak and extent of significant activation for the same contrast in a matched group of healthy controls (G.E. Strangman, R. Goldstein, T.M. O'Neil-Pirozzi, et al, unpublished data, 2008). As secondary analyses, we considered the average percent signal change within the symmetric right DLPFC and VLPFC regions as well as 2 other regions that were identified as differentially activated in this task for controls versus participants with TBI (fig 1).

  • View full-size image.
  • Fig 1. 

    ROIs for our prediction analyses overlaid on average anatomic scans from our 54 TBI participants. Primary ROIs are in white, and secondary ROIs are in black. Abbreviations: L, left; R, right.

Multiple Imputation 

Although no data were missing for our outcome variables, data were missing for our fMRI predictor (10 individual runs with motion artifacts) and severity of TBI predictor (n=5: records destroyed, 3; insufficient documentation, 2). Multiple imputation was used to remedy this.66 Although multiple imputation assumes that the data are missing at random, an untestable assumption, we believe that this is a reasonable assumption here; in particular, there is little reason to believe that the missingness is related to the actual value. The assumption is made more plausible by including enough variables in the imputation model (in our case, 15 variables for 108 observations, 2 runs each for 54 people).67 Twenty imputations were made through the multiple-image coordinate extraction algorithm, and Royston's software68 in Statab was used to combine the 20 resulting models for each analysis.66

Model Construction 

Regression models for each of the 6 regions of interest (ROIs) were initially fit with all predictor variables (age, education, moderate TBI, severe TBI, pretest HVLT, average fMRI percentage change). Terms were removed if found clearly nonsignificant (P>0.1) unless they comprised a subcomponent of a variable that was significant (eg, a linear term for which a nonlinear term was significant or the components of severity). Candidate final models were checked for nonadditivity, nonnormality of residuals, and quadratic or cubic components. One model was found to require modification during these checks, as discussed later.

Back to Article Outline

Results 

Demographic and Cognitive Characterization 

Demographic variables and injury severity appear in Table 1, Table 2, respectively. In our sample, nearly 45% of participants suffered severe injuries. Cognitive test results for all 3 time points are shown in table 3. Participants showed significant improvement between the pretest and posttest 2 for both the primary (HVLT-R) and secondary (RBMT-II) outcome measures (table 3, rightmost column).

Table 3. Neuropsychologic Assessments of Study Participants
AssessmentMean Pretest ± SDMean Posttest 1 ± SDMean Posttest 2 ± SDPretest vs Posttest 2 (% change)Pretest vs Posttest 2 (T value)
HVLT immediate recall21.6±5.424.3±5.725.4±5.4186.4⁎
HVLT delayed recall5.7±3.48.1±3.28.2±3.0448.5⁎
RBMT (total)16.1±4.218.2±4.918.5±4.2155.2⁎
TMT-A (s)51.4±5636.8±15.142.0±47.7−18−4.3⁎
TMT-B (s)116.8±104.794.4±48.488.5±67.4−24−4.2⁎
Digit span (combined)13.4±4.514.1±4.915.1±3.9134.3⁎
BNT13.4±2.2NANANANA
BDAE17.8±5.7NANANANA

Abbreviations: BDAE, Boston Diagnostic Aphasia Examination; BNT, Boston Naming Test 2nd edition; NA, not applicable.

⁎P<.001 for each test.

Prediction Results 

We fit a total of 12 models: one for HVLT-R and one for RBMT-II in each of our 6 ROIs. The HVLT-R models in DLPFC and VLPFC were our 2 primary models. None of the final models revealed significant effects of age, education, or having a moderate versus mild injury (P>.25 for each, suggesting the power to detect differences was not a major factor). In addition, every model included a significant positive loading on preintervention HVLT-R (or RBMT-II), indicating, as expected, that higher pretest scores on the outcome variable were associated with higher posttest scores.

Of the remaining 2 prediction variables (severe TBI and fMRI), models for both primary ROIs revealed a significant predictive contribution of having a severe relative to a mild injury (left VLPFC, P=.049; left DLPFC, P=.041) (table 4). The identical negative loadings in each case indicated that severe injuries were associated with a 1.4-word decrease in postintervention HVLT-R scores. The fMRI variable was not found linearly predictive for either primary region. However, a significant nonlinear (quadratic) effect of the fMRI variable was found for the left VLPFC ROI (that was the only significant finding in all our check analyses). In particular, the nonlinear fMRI effect was manifest as an inverted-U quadratic centered on an fMRI value of 0.1% change. Thus, increases or decreases in the left VLPFC fMRI signal from the value of 0.1% during the directed condition were associated with lower posttest HVLT-R scores. No other significant quadratic (or cubic) relationships were revealed.

Table 4. Final Prediction Model Results: Primary ROIs
Model VariableLeft VLPFCLeft DLPFC
Coef (SE)95% CIT(P)Coef (SE)95% CIT(P)
Pre-HVLT-R0.51(0.11)0.29to0.734.60(<.001)0.51(0.10)0.31to0.715.00(<.001)
fMRI7.60(4.40)−1.3to16.61.70(.092)NANANA
fMRI × fMRI−36.10(12.60)−61.2to−10.6−2.86(.007)NANANA
Moderate TBI−0.57(0.56)−1.7to−0.55−1.00(.31)−0.65(0.58)−1.8to0.51−1.10(.26)
Severe TBI−1.40(0.70)−2.8to−0.006−2.00(.049)−1.40(0.70)−2.9to−0.065−2.10(.041)
Intercept6.10(0.80)4.5to7.87.60(<.001)6.20(0.80)4.4to7.97.40(<.001)

Abbreviations: CI, confidence interval; Coef, coefficient; NA, not applicable; SE, standard error.

When analyzing our 4 secondary ROIs, none revealed a significant predictive contribution of fMRI over and above the other variables. In secondary analyses of RBMT-II, only the pretest RBMT-II score was found predictive of outcome.

Back to Article Outline

Discussion 

In this study, we assessed the ability of fMRI to help predict rehabilitation outcomes from a group-based strategic memory intervention in memory-impaired participants with TBI. We hypothesized that the particular left DLPFC and left VLPFC regions that were strongly activated by control participants in a strategic memory task would be important for successful strategic memory rehabilitation. We further presumed that the magnitude of activation in these regions during a verbal encoding task involving a semantic clustering strategy would help predict the outcome of a memory rehabilitation intervention focused on internal memorization strategies. The magnitude of activation in the left VLPFC did indeed provide significant predictive value for rehabilitation outcomes, even after adjusting for other typical predictors of long-term outcome. This relationship was an inverted-U quadratic, indicating that there was an optimal level of activation in this region, with either higher or lower levels of activation associated with poorer outcomes. Again, this finding is equivalent to finding a quadratic relationship between left VLPFC activation and the change in HVLT pre- versus postintervention.

The VLPFC has previously been associated with various strategic processes, including semantic retrieval,69 cognitive control,14 and processing competing memories.70 Based on these previous findings, we interpret our quadratic finding as follows. There is an optimal level and/or type of strategic processing and an associated optimal level of activation in this region, which was represented by participants who activated in the midrange with relative success on memory testing. The remainder of our sample of subjects with TBI included underactivators and overactivators of left VLPFC, either because of the type or amount of strategic processing they brought to bear. Subjects who underactivated the VLPFC may have been unable to do so either as a result of focal structural injury to gray or white matter in that region, because of damage or disruption in regions projecting to the VLPFC, or because they engaged substantially different networks. People who overactivated the VLPFC, on the other hand, presumably had an intact VLPFC and engaged in effortful utilization of this area, but damage elsewhere did not allow this effortful utilization to be realized as an improvement on testing. The overactivation may be related to unsuccessful efforts to compensate for this damage in other areas. This interpretation suggests a specific prediction, namely that the anatomic and functional connectivity associated with TBI are differently distributed in under- versus overactivators. We plan to test this hypothesis through structural and/or diffusion MRI studies when also combined with functional activation prediction studies.

In general, fMRI activation may provide unique information for outcome prediction by identifying regions that are not functioning within normative parameters, regardless of behavioral performance on baseline memory testing. For example, assume a person with TBI reveals near-normal performance during strategic verbal learning. This could arise if (1) there is little or no damage to the network relevant to strategic verbal learning or (2) there is damage to the underlying network, but the individual engaged the network differently (in magnitude or number of areas activated) or used different compensatory strategies that engage alternate brain networks. fMRI may help differentiate these cases, although doing so may require more sophisticated models that incorporate multiple nodes at a time from the cerebral network activated by the task (ie, multiple fMRI predictors in a model).

Initial injury severity was also a contributor to outcomes prediction, wherein participants with severe injuries scored significantly lower on outcomes than those with mild injuries. Such a significant effect of severity is consistent with several previous studies54, 56, 59, 61, 62 evaluating more general outcomes such as return to work or level of disability. However, injury severity is also typically found to be a less sensitive predictor of outcome at times more remote from injury.71, 72, 73 Thus, it is intriguing that an acute severity measure provided a predictive value even 11 years after injury (on average), at least when measuring rehabilitation outcomes with this particular intervention. Significant effects were not observed for moderate versus mild severity, although differences in group sizes across severity likely provided a lower power to detect differences in that group.

It is surprising that age did not contribute to the prediction of rehabilitation outcome given its frequent association with outcome in other studies.54, 55, 56, 57, 58, 59 Again, however, most of these studies assessed a more general outcome scenario rather than the prediction of outcome in a particular cognitive domain or in response to a specific intervention. This finding suggests that although the prognosis of older individuals sustaining a recent TBI is usually considered poorer, these same individuals may, at a later date, benefit just as much from a structured rehabilitation intervention as younger patients.42 Further research is needed in this area.

For our secondary outcome variable, namely, the RBMT-II, none of our predictors were significant except for the pretest RBMT-II score. The lack of a significant fMRI contribution in predicting RBMT-II outcomes could arise from many sources. For example, there was a smaller percentage improvement in RBMT-II scores relative to HVLT-R scores postintervention, making the change more difficult to predict. In addition, our choice of ROIs and probe task may have adversely affected the predictability of RBMT-II outcomes. In particular, the RBMT-II assesses memory more broadly than the HVLT-R.74 Although never tested through FNI, the RBMT-II tasks would therefore be expected to engage at least partly separable brain networks because it includes spatial memory, declarative memory, and working memory challenges. We selected our ROIs based only on a semantic clustering task, and we also used a semantic clustering task as a functional probe of the underlying systems. Although it is a reasonable starting point, this approach may have been less optimized for predicting more general RBMT-II outcome. This suggests the value of a brain-wide search for other ROIs (or multiple ROIs) or a more consonant probe task to optimize predictive power.

Study Limitations 

Finally, we consider the limitations of this study. We recognize that the contribution of fMRI signal change to our prediction model may be small clinically speaking. As mentioned, we view this study as a first step and that several factors must be addressed that can improve predictive power. For example, our focused fMRI probe task (semantic verbal list learning) only begins to encompass the aspects of strategic, semantic, or executive manipulation of memory contents that were addressed in the intervention. Similarly, our primary outcome (HVLT-R) does not encompass a wide range of strategy- or memory-related capabilities and capacities. Although our directed encoding condition during fMRI and the HVLT-R outcome are a reasonable starting point and a suitably matched pair, optimizations of both for predicting rehabilitation outcomes are certainly possible and should be investigated.

We also note that we chose a priori a relatively simple measure of brain function (average percentage signal change in directed vs fixation) over relatively broad ROIs. Although this approach proved modestly effective here, other cerebral activation measures such as maximal activation within the ROI, differential activation of directed versus spontaneous, percent of healthy control activity levels, or even scanning during retrieval rather than encoding may provide even greater predictive value by more directly probing the (thus far inadequately delineated) regions contributing to memory difficulties post-TBI. In addition, although the lateral prefrontal cortex has not been well differentiated into separate functional subregions, it is quite possible that the large ROIs used here encompass multiple distinct regions of heterogeneous processing.75 Computation of an average percentage signal change over large ROIs may have washed out regionally stronger signals. And, as also mentioned earlier, functional disruptions can be caused by damage to multiple places in a neuronal network and, conversely, damage to a single region can affect the functioning of multiple network components. Thus, the use of multiple fMRI predictor variables may also improve predictive power. This strongly suggests the need for future work to examine the regional specificity of predictive findings and the importance of multiple cerebral measures to enhance prediction.

Finally, regarding the nonsignificant predictors, it is possible that our study had insufficient power to detect these effects (eg, age, education, moderate severity). The coefficient estimates were small, however, not reaching a 1-word change in HVLT-R score across 16 years of education or across 30 years of age. Thus, even if significant, the effects of these variables would not generally be deemed important in this outcome prediction context.

Back to Article Outline

Conclusions 

We have shown that using the regional fMRI percentage of signal change obtained from a functional probe task can provide significant, additional predictive power regarding cognitive rehabilitation outcomes over and above standard predictor variables. Several procedural optimizations, including the probe task during scanning, the particular measure of neural activity, selection of ROIs for prediction, the nature and content of the cognitive intervention, and selection of the outcome measure, all remain to be performed, and such optimizations could reasonably be expected to improve predictive value. Nevertheless, our findings provide considerable promise for using functional neuroimaging to help select rehabilitation procedures and guide the rehabilitation process after TBIs.

Suppliers

Back to Article Outline

Acknowledgments 

We thank Dana Christina, MS, The Commonwealth of Massachusetts Brain Injury and Statewide Specialized Community Services Department, the Brain Injury Association of Massachusetts, and Community Rehab Care for their help and support.

Back to Article Outline

References 

  1. Curtiss G, Vanderploeg RD, Spencer J, Salazar AM. Patterns of verbal learning and memory in traumatic brain injury. J Int Neuropsychol Soc. 2001;7:574–585
  2. Vanderploeg RD, Crowell TA, Curtiss G. Verbal learning and memory deficits in traumatic brain injury: encoding, consolidation, and retrieval. J Clin Exp Neuropsychol. 2001;23:185–195
  3. McKinlay W, Watkis AJ. Cognitive and behavioral effect of brain injury. In:  Rosenthal M,  Bond ER,  Griffith MR,  Dou Miller J editor. Rehabilitation of the adult and child with traumatic brain injury. Philadelphia: FA Davis; 1999;p. 74–116
  4. Bate AJ, Mathias JL, Crawford JR. Performance on the Test of Everyday Attention and standard tests of attention following severe traumatic brain injury. Clin Neuropsychol. 2001;15:405–422
  5. Busch RM, McBride A, Curtiss G, Vanderploeg RD. The components of executive functioning in traumatic brain injury. J Clin Exp Neuropsychol. 2005;27:1022–1032
  6. Cicerone K, Levin H, Malec J, Stuss D, Whyte J. Cognitive rehabilitation interventions for executive function: moving from bench to bedside in patients with traumatic brain injury. J Cogn Neurosci. 2006;18:1212–1222
  7. McDowell S, Whyte J, D'Esposito M. Working memory impairments in traumatic brain injury: evidence from a dual-task paradigm. Neuropsychologia. 1997;35:1341–1353
  8. Levine B, Dawson D, Boutet I, Schwartz ML, Stuss DT. Assessment of strategic self-regulation in traumatic brain injury: its relationship to injury severity and psychosocial outcome. Neuropsychology. 2000;14:491–500
  9. Cazalis F, Azouvi P, Sirigu A, Agar N, Burnod Y. Script knowledge after severe traumatic brain injury. J Int Neuropsychol Soc. 2001;7:795–804
  10. Ryan TV, Sautter SW, Capps CF, Meneese W, Barth JT. Utilizing neuropsychological measures to predict vocational outcome in a head trauma population. Brain Inj. 1992;6:175–182
  11. Ryan TV. Short-term remediation of memory deficits in moderate to severe closed head injuries. [PhD dissertation] San Diego: California School of Professional Psychology; 1986;
  12. Kennedy M. Managing memory and metamemory impairment in individuals with traumatic brain injury. AHSA Lead. 2006;11:8–934-6
  13. Stuss DT, Gow CA. “Frontal dysfunction” after traumatic brain injury. Neuropsychiatry Neuropsychol Behav Neurol. 1992;5:272–282
  14. Bunge SA, Burrows B, Wagner AD. Prefrontal and hippocampal contributions to visual associative recognition: interactions between cognitive control and episodic retrieval. Brain Cogn. 2004;56:141–152
  15. Badre D, Wagner AD. Selection, integration, and conflict monitoring; assessing the nature and generality of prefrontal cognitive control mechanisms. Neuron. 2004;41:473–487
  16. Gershberg FB, Shimamura AP. Impaired use of organizational strategies in free recall following frontal lobe damage. Neuropsychologia. 1995;33:1305–1333
  17. Shimamura AP, Janowsky JS, Squire LR. What is the role of frontal lobe damage in memory disorders?. In:  Levin HS,  Eisenberg HM,  Benton AL editor. Frontal lobe function and dysfunction. New York: Oxford Univ Pr; 1991;p. 173–195
  18. Strangman G, O'Neil-Pirozzi TM, Burke D, et al. Functional neuroimaging and cognitive rehabilitation for people with traumatic brain injury. Am J Phys Med Rehabil. 2005;84:62–75
  19. Ruff RM, Baser CA, Johnston JW, et al. Neuropsychological rehabilitation: an experimental study with head-injured patients. J Head Trauma Rehabil. 1989;4:20–36
  20. Lombardi WJ, Andreason PJ, Sirocco KY, et al. Wisconsin Card Sorting Test performance following head injury: dorsolateral fronto-striatal circuit activity predicts perseveration. J Clin Exp Neuropsychol. 1999;21:2–16
  21. Kirkby BS, Van Horn JD, Ostrem JL, Weinberger DR, Berman KF. Cognitive activation during PET: a case study of monozygotic twins discordant for closed head injury. Neuropsychologia. 1996;34:689–697
  22. McAllister TW, Saykin AJ, Flashman LA, et al. Brain activation during working memory 1 month after mild traumatic brain injury: a functional MRI study. Neurology. 1999;53:1300–1308
  23. McAllister TW, Flashman LA, Sparling MB, Saykin AJ. Working memory deficits after traumatic brain injury: catecholaminergic mechanisms and prospects for treatment—a review. Brain Inj. 2004;18:331–350
  24. Christodoulou C, DeLuca J, Ricker JH, et al. Functional magnetic resonance imaging of working memory impairment after traumatic brain injury. J Neurol Neurosurg Psychiatry. 2001;71:161–168
  25. Ricker JH, Müller RA, Zafonte RD, Black KM, Millis SR, Chugani H. Verbal recall and recognition following traumatic brain injury: a [0-15]-water positron emission tomography study. J Clin Exp Neuropsychol. 2001;23:196–206
  26. Berg I, Koning-Haanstra M, Deelman B. Long term effects of memory rehabilitation (A controlled study). Neuropsychol Rehabil. 1991;1:97–111
  27. Kaschel R, Sala SD, Cantagallo A, Fahlböck A, Laaksonen R, Kazen M. Imagery mnemonics for the rehabilitation of memory: a randomised group controlled trial. Neuropsychol Rehabil. 2002;12:127–153
  28. Ryan TV, Ruff RM. The efficacy of structured memory retraining in a group comparison of head trauma patients. Arch Clin Neuropsychol. 1988;3:165–179
  29. Hewitt J, Evans JJ, Dritschel B. Theory driven rehabilitation of executive functioning: improving planning skills in people with traumatic brain injury through the use of an autobiographical episodic memory cueing procedure. Neuropsychologia. 2006;44:1468–1474
  30. Povlishock JT, Katz DI. Update of neuropathology and neurological recovery after traumatic brain injury. J Head Trauma Rehabil. 2005;20:76–94
  31. Consensus conference. Rehabilitation of persons with traumatic brain injury (NIH consensus development panel on rehabilitation of persons with traumatic brain injury). JAMA. 1999;282:974–983
  32. O'Shanick GJ, O'Shanick AM. Personality and intellectual changes. In:  Silver JM,  Yudofsky SC,  Hales RE editor. Neuropsychiatry of traumatic brain injury. Washington (DC): Am Psychiatric Pr; 1994;p. 163–188
  33. Bajo A, Fleminger S. Brain injury rehabilitation: what works for whom and when?. Brain Inj. 2002;16:385–395
  34. Ylvisaker M, Hanks R, Johnson-Greene D. Perspectives on rehabilitation of individuals with cognitive impairment after brain injury: rationale for reconsideration of theoretical paradigms. J Head Trauma Rehabil. 2002;17:191–209
  35. Zencius A, Wesolowski MD, Krankowski T, Burke WH. Memory notebook training with traumatically brain-injured clients. Brain Inj. 1991;5:321–325
  36. Laatsch LK, Thulborn KR, Krisky CM, Shobat DM, Sweeney JA. Investigating the neurobiological basis of cognitive rehabilitation therapy with fMRI. Brain Inj. 2004;18:957–974
  37. Musso M, Weiller C, Kiebel S, Muller SP, Bulau P, Rijntjes M. Training-induced brain plasticity in aphasia. Brain. 1999;122(Pt 9):1781–1790
  38. Small SL, Hlustik P, Noll DC, Genovese C, Solodkin A. Cerebellar hemispheric activation ipsilateral to the paretic hand correlates with functional recovery after stroke. Brain. 2002;125(Pt 7):1544–1557
  39. Loubinoux I, Dechaumont-Palacin S, Castel-Lacanal E, et al. Prognostic value of FMRI in recovery of hand function in subcortical stroke patients. Cereb Cortex. 2007;17:2980–2987
  40. Hoeft F, Meyler A, Glover GH, et al. Prediction of children's reading skills using behavioral, functional and structural neuroimaging measures. Behavi Neurosci. 2007;121:602–613
  41. Savage CR, Deckersbach T, Heckers S, et al. Prefrontal regions supporting spontaneous and directed application of verbal learning strategies: evidence from PET. Brain. 2001;124(Pt 1):219–231
  42. Logan JM, Sanders AL, Snyder AZ, Morris JC, Buckner RL. Under-recruitment and nonselective recruitment: dissociable neural mechanisms associated with aging. Neuron. 2002;33:827–840
  43. Cicerone KD, Dahlberg C, Kalmar K, et al. Evidence-based cognitive rehabilitation: recommendations for clinical practice. Arch Phys Med Rehabil. 2000;81:1596–1615
  44. Brandt J, Benedict RH. Hopkins Verbal Learning Test−Revised. Psychological Assessment Resources; 2001;
  45. Preece MH, Geffen GM. The contribution of pre-existing depression to the acute cognitive sequelae of mild traumatic brain injury. Brain Inj. 2007;21:951–961
  46. Bazarian JJ, Wong T, Harris M, Leahey N, Mookerjee S, Dombovy M. Epidemiology and predictors of post-concussive syndrome after minor head injury in an emergency population. Brain Inj. 1999;13:173–189
  47. Barr WB. Neuropsychological testing of high school athletes (Preliminary norms and test-retest indices). Arch Clin Neuropsychol. 2003;18:91–101
  48. Wilson B, Cockburn J, Baddeley A, Hiorns R. The Rivermead Behavioural Memory Test-II Supplement Two. Bury St. Edmunds: Thames Valley Test; 2003;
  49. Malec J, Zweber B, DePompolo R. The Rivermead Behavioural Memory Test, laboratory neurocognitive measures, and everyday functioning. J Head Trauma Rehabil. 1990;5:60–68
  50. Wilson B, Cockburn J, Baddeley A, Hiorns R. The development and validation of a test battery for detecting and monitoring everyday memory problems. J Clin Exp Neuropsychol. 1989;11:855–870
  51. Marchini JL, Ripley BD. A new statistical approach to detecting significant activation in functional MRI. Neuroimage. 2000;12:366–380
  52. Crinion J, Ashburner J, Leff A, Brett M, Price C, Friston K. Spatial normalization of lesioned brains: performance evaluation and impact on fMRI analyses. Neuroimage. 2007;37:866–875
  53. Benedict R, Schretlen D, Groninger L, Brandt J. Hopkins verbal learning test-revised: normative data and analysis of inter-form and test retest reliability. Clin Neuropsychol. 1998;12:43–55
  54. Katz DI, Alexander MP. Traumatic brain injury (Predicting course of recovery and outcome for patients admitted to rehabilitation). Arch Neurol. 1994;51:661–670
  55. Ruff RM, Marshall LF, Crouch J, et al. Predictors of outcome following severe head trauma: follow-up data from the Traumatic Coma Data Bank. Brain Inj. 1993;7:101–111
  56. Brown AW, Malec JF, McClelland RL, Diehl NN, Englander J, Cifu DX. Clinical elements that predict outcome after traumatic brain injury: a prospective multicenter recursive partitioning (decision-tree) analysis. J Neurotrauma. 2005;22:1040–1051
  57. Millis SR, Rosenthal M, Novack TA, et al. Long-term neuropsychological outcome after traumatic brain injury. J Head Trauma Rehabil. 2001;16:343–355
  58. Keyser-Marcus LA, Bricout JC, Wehman P, et al. Acute predictors of return to employment after traumatic brain injury: a longitudinal follow-up. Arch Phys Med Rehabil. 2002;83:635–641
  59. Temkin NR, Holubkov R, Machamer JE, Winn HR, Dikmen SS. Classification and regression trees (CART) for prediction of function at 1 year following head trauma. J Neurosurg. 1995;82:764–771
  60. Greenspan AI, Wrigley JM, Kresnow M, Branche-Dorsey CM, Fine PR. Factors influencing failure to return to work due to traumatic brain injury. Brain Inj. 1996;10:207–218
  61. Sherer M, Sander AM, Nick TG, High WM, Malec JF, Rosenthal M. Early cognitive status and productivity outcome after traumatic brain injury: findings from the TBI model systems. Arch Phys Med Rehabil. 2002;83:183–192
  62. van Zomeren AH, van den Burg W. Residual complaints of patients two years after severe head injury. J Neurol Neurosurg Psychiatry. 1985;48:21–28
  63. Dikmen SS, Machamer JE, Powell JM, Temkin NR. Outcome 3 to 5 years after moderate to severe traumatic brain injury. Arch Phys Med Rehabil. 2003;84:1449–1457
  64. Reber PJ, Siwiec RM, Gitelman DR, Parrish TB, Mesulam MM, Paller KA. Neural correlates of successful encoding identified using functional magnetic resonance imaging. J Neurosci. 2002;22:9541–9548
  65. Miotto EC, Savage CR, Evans JJ, et al. Bilateral activation of the prefrontal cortex after strategic semantic cognitive training. Hum Brain Mapp. 2006;27:288–295
  66. Little R, Rubin D. Statistical analysis with missing data. 2nd ed.. New York: Wiley; 2002;
  67. Schafer JL. Analysis of incomplete multivariate data. London: Chapman & Hall; 1997;
  68. Royston P. Multiple imputation of missing values: update of ice. Stata J. 2005;5:527–536
  69. Bunge SA, Kahn I, Wallis JD, Miller EK, Wagner AD. Neural circuits subserving the retrieval and maintenance of abstract rules. J Neurophysiol. 2003;90:3419–3428
  70. Kuhl BA, Dudukovic NM, Kahn I, Wagner AD. Decreased demands on cognitive control reveal the neural processing benefits of forgetting. Nat Neurosci. 2007;10:908–914
  71. Corrigan JD, Smith-Knapp K, Granger CV. Outcomes in the first 5 years after traumatic brain injury. Arch Phys Med Rehabil. 1998;79:298–305
  72. Novack TA, Bush BA, Meythaler JM, Canupp K. Outcome after traumatic brain injury: pathway analysis of contributions from premorbid, injury severity, and recovery variables. Arch Phys Med Rehabil. 2001;82:300–305
  73. Bush BA, Novack TA, Malec JF, Stringer AY, Millis SR, Madan A. Validation of a model for evaluating outcome after traumatic brain injury. Arch Phys Med Rehabil. 2003;84:1803–1807
  74. Strauss E, Sherman E, Spreen O. A compendium of neurosychological tests: administration, norms, and commentary. 3rd ed.. New York: Oxford Univ Pr; 2006;
  75. Badre D, Poldrack RA, Paré-Blagoev EJ, Insler RZ, Wagner AD. Dissociable controlled retrieval and generalized selection mechanisms in ventrolateral prefrontal cortex. Neuron. 2005;47:907–918
  • a Siemens, Citicorps Center, 153 E 53rd St, New York, NY 10022-4611.
  • b Version 9.2; Stata, 4905 Lakeway Dr, College Station, TX 77845.

 Supported by National Institute on Disability and Rehabilitation Research (grant no. H133A020513) and the National Institute of Neurological Disorders and Stroke (grant no. K25-NS046554).No commercial party having a direct financial interest in the results of the research supporting this article has or will confer a benefit upon the authors or upon any organization with which the authors are associated.

PII: S0003-9993(08)00124-X

doi:10.1016/j.apmr.2008.02.011

Archives of Physical Medicine and Rehabilitation
Volume 89, Issue 5 , Pages 974-981, May 2008

Article Tools

* Image Usage & Resolution