S. Laureys et al. (Eds.)Progress in Brain Research, Vol. 177ISSN 0079-6123Copyright r 2009 Elsevier B.V. All rights reserved

CHAPTER 6

Predictors of short-term outcome in brain-injuredpatients with disorders of consciousness$

J. Whyte1,�, O. Gosseries2, I. Chervoneva3, M.C. DiPasquale1, J. Giacino4, K. Kalmar4,D.I. Katz5, P. Novak6, D. Long7, N. Childs8, W. Mercer8, P. Maurer9 and B. Eifert9

1Moss Rehabilitation Research Institute/Albert Einstein Healthcare Network, Philadelphia, PA, USA2Coma Science Group, Cyclotron Research Center, University of Liege, Liege, Belgium3Division of Biostatistics, Department of Pharmacology and Experimental Therapeutics,

Thomas Jefferson University, Philadelphia, PA, USA4JFK Johnson Rehabilitation Institute, Edison, NJ, USA

5Brain Injury Program, Braintree Rehabilitation Hospital, Braintree, MA; Department of Neurology,Boston University School of Medicine, Boston, MA, USA

6Sunnyview Hospital and Rehabilitation Center, Schenectady, NY, USA7Bryn Mawr Rehab Hospital, Malvern, PA, USA

8Texas NeuroRehab Center, Austin, TX, USA9Fachkrankenhaus Neresheim Hospital, Neresheim, Germany

Abstract:Objectives: To investigate predictors of recovery from the vegetative state (VS) and minimally consciousstate (MCS) after brain injury as measured by the widely used Disability Rating Scale (DRS) and toexplore differences in rate of recovery and predictors of recovery during inpatient rehabilitation inpatients with non-traumatic (NTBI) and traumatic brain injury (TBI).

Design: Longitudinal observational cohort design and retrospective comparison study, in which an initialDRS score was collected at the time of study enrollment. Weekly DRS scores were recorded until dischargefrom the rehabilitation center for both NTBI and TBI patients.

Setting: Seven acute inpatient rehabilitation facilities in the United States and Europe with specializedprograms for VS and MCS patients (the Consciousness Consortium).

Participants: One hundred sixty-nine patients with a non-traumatic (N ¼ 50) and a traumatic (N ¼ 119)brain injury who were in the VS or MCS states.

Interventions: Not applicable.

Main Outcome Measures: DRS score at 13 weeks after injury; change in DRS score over 6 weeks post-admission; and time until commands were first followed (for patients who did not show command-following at or within 2 weeks of admission).

$Both J. Whyte and O. Gosseries have contributed equallyto this study.

�Corresponding author.Tel.: 00 1 215 663-6872; Fax: 00 1 215 663-6113;E-mail: JWHYTE@einstein.edu

DOI: 10.1016/S0079-6123(09)17706-3 63

64

Results: Both time between injury and enrollment and DRS score at enrollment were significant predictorsof DRS score at week 13 post-injury but the main effect of etiology only approached significance. Etiologywas however a significant predictor of the amount of recovery observed over the 6 weeks followingenrollment. Time between injury and enrollment was also a good predictor of this outcome, but not DRSscore at enrollment. For the time until commands were first followed, patients with better DRS scores atenrollment, and those with faster early rates of change recovered command following sooner than thosewith worse DRS scores or slower initial rates of change. The etiology was not a significant predictor for thislast outcome. None of these predictive models explained sufficient variance to allow their use in individualclinical decision making.

Conclusions: Time post-injury and DRS score at enrollment are predictors of early recovery amongpatients with disorders of consciousness, depending on the outcome measure chosen. Etiology was also asignificant predictor in some analyses, with traumatically injured patients recovering more than those withnon-traumatic injuries. However, the hypothesized interaction between etiology and time post-injury didnot reach significance in any of the analyses suggesting that, within the time frame studied, the decline inprognosis with the passage of time was similar in the two groups.

Keywords: brain injuries; minimally conscious state; vegetative state; Disability Rating Scale; followingcommand; prognosis; consciousness

Introduction but when the model is being used to allocateclinical services, one may be interested primarily

Outcome prediction is a frequent topic in the in the outcome within the time frame that thoseliterature on neurologic recovery and rehabilita- services will be provided.tion. However, one may have several different Prediction of outcome among patients withpurposes in mind in outcome prediction and each disorders of consciousness (DOC) is still difficultof these purposes places different performance to establish individually. Moreover, most prog-requirements on the predictive model. Relatively nostic studies have begun on the day of injurygross aggregate prediction of rates of recovery when the diagnoses of vegetative (VS) andmay suffice for the purpose of planning healthcare minimally conscious states (MCS) are not yetservices, estimating costs, or generating payment defined, and have studied the full range of injuryschemes. Similar gross aggregate models may severity. This provides little guidance to clinicianshighlight predictor variables that may have who see patients who have evolved from comatheoretical interest as possible causal factors in into the VS or MCS, and who wish to assess therecovery. A much more demanding use of out- likelihood of further progress, to determine thecome prediction is to assist in the healthcare appropriate level of treatment intensity, and todecision making for individual patients. Here one provide guidance to caregivers in their decisionmight wish to avoid ‘‘wasting’’ resources on making.someone who will not show substantial recovery, It is known that among patients with DOC oneand to ensure that someone with good recovery month after injury, those who show some minimalpotential receives services that will optimize that signs of consciousness have a better chance ofrecovery. In this context, even a relatively recovery than patients who are still in a VS at thataccurate aggregate model may make inaccurate time, and the earlier the return of consciousness ispredictions about substantial numbers of indivi- detected, the better is the outcome (Giacino anddual cases. Outcome prediction may also differ in Zasler, 1997; Giacino and Whyte, 2005; Whytethe time frame of interest. In many cases, the et al., 2005; Giacino and Kalmar, 1997). The‘‘final outcome’’ is of greatest interest to predict, etiology is also a relevant predictor of recovery.

65

Traumatic brain injury (TBI) tends to have abetter outcome than non-traumatic etiology(NTBI) (especially anoxia) (The Multi-SocietyTask Force on PVS, 1994b). Moreover, therecovery phase lasts longer for a traumaticetiology: it has been suggested that the termpermanent vegetative state should not be applieduntil 1 year after traumatic injury whereas for anon-traumatic injury, this diagnosis may beapplied after only 3 months (The Multi-SocietyTask Force on PVS, 1994a, b). Note that the termpermanent implies zero probability of recoveryand can therefore give rise to serious decisionsabout the cessation of medication and nutrition.

Potential recovery is also linked to the location,extent, and nature of the brain damage as well asto the condition of the brain before the injury.Young age of the patient and the absence ofmedical history (such as alcoholism, drug use, ormental illness) lead to a better outcome (TheMulti-Society Task Force on PVS, 1994b; Laureyset al., 2001). For patients with DOC of traumaticorigin, the Disability Rating Scale (DRS) at 16weeks post-injury, and the time at which com-mands were first followed, during the acuterehabilitation hospitalization, were related to theDRS score at rehabilitation admission, the timebetween injury and admission, and rate of DRSchange during the first 2 weeks of rehabilitation(Whyte et al., 2005). New assessment methods,such as event-related potential (ERP) techniques,and evaluation with functional imaging modalitiessuch as positron emission tomography (PET) andfunctional magnetic resonance imaging (fMRI)scanning, offer promise in improving the precisionof prognostic prediction, since they may helpdistinguish among patients with different neuro-physiologic profiles (which confer different prog-noses) at a time when behavioral assessments areat floor for all of them (Di et al., 2008; Owen et al.,2006; Schnakers et al., 2008; Kotchoubey, 2007).However, although these techniques appear to beable to identify a subgroup of VS patients withgreater recovery potential or to identify subtlesigns of consciousness not apparent on behavioralexamination, they have not yet been used insystematic prediction at defined time points post-injury along with already known predictors.

Research on prediction of recovery from DOCis particularly challenging to conduct, at least inthe United States, because intensive academicallyoriented healthcare services are severelyrestricted for this patient population after the firstfew weeks post-onset. This is based on a generalpessimism that meaningful recovery is unlikely,the belief that the process of rehabilitationrequires a level of voluntary participation thatsuch patients cannot meet, and the sense thatthere is little evidence that intensive rehabilitationservices can alter the outcome. Thus, such patientsare generally dispersed to family homes or non-specialized nursing care facilities soon after injury,and, accordingly, lost to involvement in long-itudinal research.

Because many of the available outcome studiesfollow a sample from the time of injury, so that alarge proportion of the sample (those with milderinjury) regain consciousness quickly, predictors ofoutcome in this rapidly recovering populationmay not apply to the sample with prolongedDOC. Other studies have followed patients withDOC for longer intervals, but typically restricttheir prediction to the return of consciousness as adichotomous variable (e.g., Multi-Society TaskForce), shedding little light on the overall level offunctional recovery. In this context, therefore, it isimportant to examine whether those patients withDOC who are available for study show sufficientrecovery during the subacute period to suggestgreater rehabilitation potential than is currentlyappreciated. In addition, if predictors of theirshort-term outcome are sufficiently accurate toguide individual service decisions, then thesecould be used to help tune rehabilitation admis-sion criteria to accept the individuals with thegreatest potential to benefit, and to avoid admit-ting individuals who will fail to make progress,and may present difficult placement problems. Inthis context, then, a number of important outcomequestions need to be addressed. (1) As a group,how much recovery do patients admitted withDOC make in the subacute period? (2) Are therevariables, available at the time that admissiondecisions are being made, that can help predictthe amount of functional recovery that will occurin the time span over which rehabilitation services

66

might be delivered? (3) Are there differences inthe factors that predict recovery for patients withtraumatic versus non-traumatic injuries duringthis interval?

We hypothesized that substantial recoverywould be seen in a large proportion of patientswho present with DOC in the first few weeks afterbrain injury. We also hypothesized, based on priorstudies and our own previous work with a puresample of individuals with traumatic brain inju-ries, that the etiology of injury (in particulartraumatic vs. non-traumatic), the time post-eventat which the patient was admitted to rehabilita-tion, and the functional level at which they wereadmitted, would all predict differences in theshort-term recovery seen over the ensuing weeks.

Methods

Participants in this research were enrolled fromthe Consciousness Consortium (CC), which con-sists of a set of facilities in the United States andEurope that have specialized programs for thecare and rehabilitation of patients with DOC, andan interest in conducting research in this area. TheCC began a longitudinal descriptive study in 1996,and reported the results of the traumatic sample(n ¼ 124) in 2005 (Whyte et al., 2005) which laidthe groundwork for a randomized controlledtreatment study currently underway (Giacino and

Table 1. Outcome variables

Whyte, 2003; Whyte, 2007). Here we report thedata for the non-traumatic sample also enrolledby CC facilities, and also analyze comparativeresults for the two subsamples.

Participants

Participants were enrolled in the study betweenDecember 1996 and June 2001 when they wereadmitted to one of the seven CC-memberrehabilitation centers. Admission criteria were asevere acquired brain injury of traumatic or non-traumatic etiology and a DRS score on admissiongreater than 15, with no more than inconsistentcommand-following. These score criteria werechosen because all patients in VS or MCS shouldhave DRS scores of at least 16, but lack ofconsistent command following helps ensure thatthose who have emerged from MCS are excluded.One hundred forty-eight (148) traumatic and 77non-traumatic patients diagnosed as vegetative orminimally conscious on admission were enteredinto the longitudinal database. However, becausespecific variables required for the analyses weremissing from some participants, the number ofparticipants included in this report is smaller (seeTable 1 for details).

Note that there is a bias of selective admissionin rehabilitation centers. Indeed, acute inpatientrehabilitation facilities tend to select, to varyingdegrees, candidates who are believed to have a

Traumatic brain injury (TBI) Non-traumatic brain injury (NTBI) Definition

No. ofsubjects

Mean/median/SD

Min./max. No. ofsubjects

Mean/median/SD

Min./max.

DRS13

ChangeDRS6

TFollow

99

99

71

18.14/20/5.65

4.96/4/4.23

62.04/30/79.72

5/28

�2/11

0/473

36

36

37

19.72/20.5/4.3

2.86/2.5/2.93

42.27/27/44.73

9/26

�1/16

0/196

DRS score at week 13post-injury[DRS score atenrollment]–[DRS score 6weeks later][Date of commandfollowing or date ofdischarge if not followingcommands]�[date of 2weeks post-admission]

67

chance of recovery and who will benefit fromintensive therapy. Admission is therefore based,at least informally, on various prognostic factorsthat are perceived to be positive indicators offunctional improvement (e.g., recent injury, pos-sible signs of consciousness, etc.). Thus, this is nota population-based study, although it is relevantto decision making in the types of facilities inwhich the study was conducted.

The study was determined by the relevantInstitutional Review Boards to be exempt fromthe need for individual informed consent becauseit involved only anonymous recording of observa-tional data but no changes in clinical care.

The Disability Rating Scale (DRS)

The DRS is a measure of impairment, disability(now referred to as ‘‘activity’’), and handicap(now referred to as ‘‘participation’’) across thespan of recovery to track an individual from comato community (Rappaport et al., 1982). The firstthree items of the DRS (‘‘Eye Opening,’’ ‘‘Com-munication Ability,’’ and ‘‘Motor Response’’)reflect impairment ratings whereas cognitiveability for ‘‘Feeding,’’ ‘‘Toileting,’’ and ‘‘Groom-ing’’ reflect the level of disability and, finally, the‘‘Level of Functioning’’ item reflects handicap, asdoes the last item, ‘‘Employability’’. The DRS isscored from 0 (no disability) to 29 (extreme VS).Note that this scale does not disentangle VS fromMCS because it was constructed before thedevelopment of the MCS criteria (Giacino et al.,2002).

Data collection

For those patients who met the enrollment criteriafor this study, demographic information, injuryhistory and early complications, and admissionDRS score were recorded. DRS scoring wasrepeated weekly as long as the patient remainedat the facility. Data were recorded on paper formsand then faxed or mailed to the data center at theMoss Rehabilitation Research Institute, wherethey were entered into a computer database. Forthese analyses, the database was queried for

demographic information (age, gender, ethnicity),the cause of injury (traumatic or non-traumaticbrain injury), the time between the injury and theadmission to the rehabilitation facility, the DRSscore on admission, the weekly DRS score untildischarge, and the time between the admissionand the first command following (if not present atadmission).

Three outcomes were addressed in the ana-lyses: the DRS score at 13 weeks after injury(DRS13), the change in DRS score over 6 weekspost-admission (ChangeDRS6), and the time untilcommands were first followed for patients whodid not show command following at or within2 weeks of admission (TFollow). Patients that didnot follow commands during admission werecensored at the discharge time. The operationaldefinition of each outcome is reported in Table 1.DRS score at 13 weeks post-injury was chosenbecause the largest sample was available at thattime and DRS score over 6 weeks post-enrollmentwas selected because it is the average length ofstay in the rehabilitation facilities. For practicalrelevance, TFollow would ideally be calculatedfrom the time of admission since a clinicianadmitting a patient wants to know whether andwhen he/she will begin to follow commandsthereafter. Moreover, calculating this index fromthe time of injury would be problematic in thissample, since many injured individuals wouldhave recovered command following much earlier,but would not be included in the sample.However, because the rate of functional improve-ment in the first 2 weeks after admission was usedas one of the predictor variables (see below fordetails), in fact we attempted to predictrecovery of command following from that pointforward.

Of the participants meeting the enrollmentcriteria, only those who had complete data forthe outcome and predictor variables were used ineach analysis (see Tables 1 and 2). This resulted inan effective sample of 135 (99TBI, 36NTBI) forthe DRS13 and ChangeDRS6 analyses and 108(71TBI, 37NTBI) for the TFollow analysis. Out ofthese 108 patients, 48 were censored at the time ofdischarge. Seventy-four (74) participants (50TBI,24NTBI) were included in all of the analyses.

68

Table 2. Predictor variables

DRS13 and 6-week change TFollow

TBI (N ¼ 99) NTBI (N ¼ 36) TBI (N ¼ 71) NTBI (N ¼ 37)

Continuous variables Mean/median/SD Mean/median/SD Mean/median/SD Mean/median/SDAge 31.58/28/14.09 40.78/39.5/16.33 28.79/26/12.54 36.89/35/14.85LogTenroll 5.55/5.39/0.55 5.44/5.51/0.49 5.59/5.39/0.59 6.05/5.73/1.29(Tenroll) (50.4/42/20.15) (45.75/45.5/14.18) (52.37/42/23.07) (115.49/53/186.19)DRSenroll 22.85/23/2.23 23.06/23/2.57 23.84/24/2.14 23.84/24/1.72

Nominal variables No. of subjects (%) No. of subjects (%) No. of subjects (%) No. of subjects (%)Gender

Male 67 (67) 15 (41.7) 54 (76.06) 13 (64.86)Female 32 (32.3) 21 (58.3) 17 (23.94) 24 (35.14)

EthnicityWhite 77 (77.8) 29 (80.5) 61 (85.91) 28 (75.68)Non-white 22 (22.2) 7 (19.4) 10 (14.09) 9 (24.32)

Abbreviations: LogTenroll: log 2 transformation of Tenroll; Tenroll: date of command following�date of enrollment; DRSenroll: DRS score atenrollment.

The characteristics of the patients in both analysessets are shown in Table 2.

Data analysis

The independent variable time to enrollment(Tenroll) was log transformed (LogTenroll), sincethe assumption of linear association with theoutcome was more appropriate on the log scale.NTBI and TBI were analyzed jointly to allowevaluation of the difference in outcomes byetiology. Different statistical models were usedfor the different outcomes. DRS at week 13 andthe change in DRS scores 6 weeks post-admissionwere analyzed on a total of 135 observations usingthe robust MM regression (Yohai, 1987), sincedistributions of residuals from the standardmultiple regression models exhibited heavy tailscompromising the normality assumption. Etiol-ogy, admission DRS, and time to enroll (log base2 transformed) as well as gender, age, andethnicity were considered as predictors of DRSat week 13 and change in DRS scores 6 weekspost-admission. The interactions betweenetiology and admission DRS, etiology and timeto enroll, and admission DRS and time to enrollwere also considered in the models. The finalmodels included the etiology, admission DRS,and time to enroll and controlled for potentiallyimportant age difference. Other demographic

variables, which were not significantly associatedwith outcome, were excluded from the models.

For the last outcome, the time until commandswere first followed, the analyses were performedon a partially overlapping sample because somepatients in the previous analyses had alreadyfollowed commands before admission andbecause some patients were admitted after 13weeks post-injury. A Cox proportional hazardsmodel was initially fitted to the time fromadmission to follow commands. Etiology, admis-sion DRS, and time to enroll (log base 2transformed) as well as gender, age, and ethnicitywere considered as predictors. The interactionsbetween etiology and admission DRS, etiologyand time to enroll, and admission DRS and timeto enroll were also considered in the model. Datafrom 169 patients were available for theseanalyses. Because the proportional hazardsassumptions were not satisfied, the 2-week rateof change in DRS was also introduced into Coxmodel, which improved the overall model fit. Inthe earlier work (Whyte et al., 2005), the 2-weekrate of change in DRS was found to be a strongpredictor of the time to follow commands in TBIpatients. Time from 2 weeks post-admission untilcommands were followed was then modeled. Thefinal Cox model was based on 108 patients whoalso had 2-week rate of change in DRS availableand did not follow commands within the first 2

69

weeks of admission. Etiology, admission DRS,and time to enroll (log base 2 transformed) as wellas gender, age, and ethnicity were considered aspredictors. The interactions between etiology andadmission DRS, etiology and time to enroll, andadmission DRS and time to enroll were alsoconsidered in the model.

Results

DRS score at week 13

None of the demographic variables was signifi-cantly associated with DRS score at week 13, butage was retained in the model because of priorresearch suggesting that age may influence thepace of neurologic recovery (Millis et al., 2001;Ritchie et al., 2000). Both time between injury andenrollment (LogTenroll) and DRS score at enroll-ment (DRSenroll) were highly significant predic-tors of DRS score at week 13 post-injury. Themain effect of etiology approached significance(difference ¼ 1.4, 95% CI: �0.2, 3.0; p ¼ 0.083).However, the interactions between etiology andDRS at enrollment and time to enrollment werenot significant. Table 3 reports the slopes for thedifferent predictors from the final model. Themodel implies that an increase of 1 point in DRSat enrollment translates on average into a 1.2point increase in DRS at week 13 (note thathigher DRS scores indicate worse outcomes).Meanwhile doubling of the time to enrollment(1 unit increase of the LogTenroll) implies a 3.7point increase in DRS at week 13. Thus, thisanalysis did not provide strong evidence for adifference in the recovery pattern between TBI

and NTBI patients during this time frame. Finally,the R2 is 0.355 for this robust regression model.

DRS score improvement over the 6 weeks post-enrollment

Etiology was a significant predictor of the amountof recovery observed over the 6 weeks followingenrollment. On average TBI patients had 2.0points (95% CI: 0.4, 3.5; p ¼ 0.011) greaterimprovement in DRS scores over the 6-weekinterval than NTBI patients. In this analysis, timeuntil enrollment, but not DRS score at enrollmentwas a significant predictor of recovery. Table 4reports the slopes for the different predictors inthe final model. The model implies that doublingof the time to enrollment (1 unit increase of thelog base 2 transformed time to enroll) impliesB1.9 point reduction in the DRS change over thisinterval. Once again, the interaction betweenetiology and DRSenroll and LogTenroll was notsignificant. Thus, although NTBI patients showedless recovery, during this interval, this lesserdegree of recovery was not accounted for by amore prominent decline in prognosis with thepassage of time. Note that the R2 for this robustregression model is only 0.094.

Time to follow commands

The final model was reduced to two significantpredictors plus etiology, because models incor-porating additional non-significant covariates didnot yield adequate goodness-of-fit test results. Asnoted in Table 5, with inclusion of the 2-week ratevariable, etiology was not a significant predictorof the time until commands were followed.

Table 3. Results for the robust regression model for DRS at week 13

Slope or differencea 95% confidence limits p-value

Lower Upper

Etiologya 1.4 �0.2 3.0 0.083LogTenroll 5.4 3.6 7.2 o0.0001DRSenroll 1.12 0.9 1.5 o0.0001Age 0.0 �0.1 0.0 0.565

aDifference between TBI and NTBI groups.

70

Table 4. Results from the robust regression model for the change in DRS scores 6 weeks post-admission

Slope or differencea 95% confidence limits p-value

Lower Upper

Etiologya 2 0.4 3.5 0.011LogTenroll �1.87 �3.12 �0.62 o0.003DRSenroll �0.14 �0.42 0.15 0.358Age 0.03 �0.01 0.08 0.184

aDifference between TBI and NTBI groups.

Table 5. Results from the Cox model for time to follow commands from 2-week admission

Hazard ratio 95% hazard ratio confidence limits p-value

Lower Upper

Etiology 1.14 0.66 1.98 0.637DRSenroll 0.87 0.77 0.99 0.0332-week rate 11.26 1.65 76.67 0.013

However, both DRSenroll and the 2-week rate ofchange were significant predictors. Patients withlower (better) DRS scores at enrollment, andthose with faster early rates of change recoveredcommand following sooner than those with higherDRS scores or slower initial rates of change.

Discussion

These results demonstrate that considerablerecovery is possible during the typical time frameof acute rehabilitation care, for both TBI andNTBI patients. Overall 83.7% of patientsimproved their DRS score by at least 1 pointover the 6 weeks of observation (84.8% of TBIand 80.5% of NTBI), and 61.1% of those whowere not following commands at admission beganto follow them prior to discharge (67.6% TBI and48.6% NTBI). How much rehabilitation servicesenhance this recovery is unknown, but thesefindings suggest that the majority of patients whoare admitted to acute rehabilitation will demon-strate meaningful recovery.

These results also confirm, in a sample of TBIand NTBI patients followed in a comparablemanner, that time between injury and enrollmentis a key predictor of recovery, with the passage of

time reducing the chances of recovery. This wastrue for DRS13 and for 6-week change, but not forthe time until commands were followed. How-ever, in the latter analysis, direct measurement ofthe rate of recovery, captured by the 2-weekchange variable, may have reduced the signifi-cance of the more indirectly predictive Tenroll

variable. The DRS score at enrollment waspredictive of the DRS score at 13 weeks post-injury, but not of the amount of recovery thatwould be seen over a defined interval, suggestingthat the admission DRS score is primarily apredictor of functional status rather than func-tional change, whereas the time until enrollment isparticularly relevant to the probability of change.DRS at enrollment was also predictive of the timeat which commands would be followed. This mayindicate that, at equivalent rates of change,patients who start at a better functional levelneed less improvement (and hence less time) toreach the criterion of command following.

The effects of etiology on outcome in this studywere more complex. NTBI patients had signifi-cantly or marginally worse outcomes in terms of6-week change and DRS13, respectively, butetiology was not a significant predictor of timeuntil commands were followed. As mentionedabove, the inclusion of the 2-week rate of change

71

in the latter model, necessitated for statisticalreasons, may have reduced the significance of lessdirect predictors such as time post-injury oretiology. Even though NTBI patients had some-what poorer outcomes than TBI patients, depend-ing on the outcome measure chosen, the specificprediction that the prognosis of NTBI patientswould decline more precipitously over time (i.e.,an interaction between etiology and LogTenroll)was not supported. This is in contrast to priorstudies suggesting that the ‘‘window of recovery’’is shorter for NTBI than for TBI patients (TheMulti-Society Task Force on PVS, 1994b). How-ever, the differential impact of time may be moredramatic in the 3–12 month range, whereas thesedata were collected primarily in the early weekspost-injury.

In the aggregate, these results confirm theimportance of etiology, initial functional status,and time since injury in determining outcome inindividuals with DOC. However, the majority ofthe variance in individual outcome remainsunaccounted for. The final models described hereaccount for approximately 35.5% of the variancein DRS13, and about 9% of the variance in 6-weekchange. Thus, these predictors cannot be used withconfidence to predict the outcomes of individualpatients or to make admission decisions, without ahigh risk of error in both directions.

This study has a number of important limita-tions. Most importantly, it was conducted on aselect referral sample, not a population-basedsample. Thus, the large proportion of patients whorecover in hours or days after injury are notincluded in the analysis. But even if one focuses onthose patients who might be considered forrehabilitation care because they are still sufferingfrom DOC several weeks post-injury, this remainsa biased sample, since it involved only thosepatients who were admitted to rehabilitationservices but not those who were not referred orwere referred but not admitted. The specificclinical factors used in making those admissiondecisions are unknown, but surely may haveincluded some subtle prognostic factors. In parti-cular, since clinicians are generally aware of themore negative prognosis of NTBI patientsreported in the literature, they may have had more

stringent admission screening of non-traumaticreferrals than of traumatic referrals. This, in turn,may have led to smaller differences in outcomebased on etiology than might be seen in a lessselected sample. This implicitly assumes, however,that some of the variance in recovery notaccounted for by the predictors used in this study,was accounted for by unmeasured variables avail-able to clinical decision makers, rather than simplybeing altogether unexplained. There is no directevidence for a more stringent admission screeningof NTBI patients since, for example, their DRSscores at enrollment were actually slightly worsethan those of the TBI patients. Finally, therelatively short-term nature of this study, con-strained by the current realities of acute inpatientrehabilitation stays in the United States, meant thata substantial number of patients were not followingcommands by the time of discharge and werecensored in the Cox analysis. Longer intervals offollow up and larger samples, particularly of thosewith non-traumatic injuries might have moreclearly informed the pattern of recovery.

Conclusion

In this selected sample of patients with DOC,referred and approved for inpatient rehabilitationadmission, significant recovery was seen over thehospital stay, with the majority of patients withboth traumatic and non-traumatic injuries demon-strating improvements in DRS scores and, amongvegetative patients, the development of commandfollowing. The time between injury and rehabili-tation admission and the DRS score at admissionwere each predictive of two of the three out-comes. Etiology was predictive of amount offunctional improvement seen over 6 weeks ofhospitalization, but less so of the DRS score at 13weeks post-injury or the time until commandswere followed. In the one model in which earlyrate of change was included, it was stronglypredictive of outcome while etiology was not,suggesting that the clinical trajectory, itself, ishighly predictive. None of the predictive outcomemodels accounted for sufficient variance to beused in individual clinical decision making.

72

Abbreviations

CC Consciousness ConsortiumChangeDRS6 change in DRS score over

6 weeks post-admissionDOC disorders of consciousnessDRS Disability Rating ScaleDRS13 Disability Rating Scale score

at 13 weeks after injuryDRSenroll Disability Rating Scale score

at enrollmentERP event-related potentialfMRI functional magnetic resonance

imagingLogTenroll log transformation of the time

between injury and enrollmentMCS minimally conscious stateNTBI non-traumatic brain injuryPET positron emission tomographyPVS persistent vegetative stateTBI traumatic brain injuryTenroll time between injury and

enrollmentTFollow time until commands were first

followed for patients who did notshow command following at orwithin 2 weeks of admission

VS vegetative state

Acknowledgments

We would like to thank Gemma Baldon andDavid Kennedy for database management. Thisstudy was supported in part by the Albert EinsteinSociety, and the Irving I. and Felicia F. RubinFamily Foundation. OG was funded by theFonds de la Recherche Scientifique (ResearchFellow), European Commission, and the James S.McDonnell Foundation.

References

Di, H., Boly, M., Weng, X., Ledoux, D., & Laureys, S. (2008).Neuroimaging activation studies in the vegetative state:Predictors of recovery? Clinical Medicine, 8, 502–507.

Giacino, J., & Kalmar, K. (1997). The vegetative and minimallyconscious states: A comparison of clinical features andfunctional outcome. The Journal of Head Trauma Rehabi-litation, 12(4), 36–51.

Giacino, J., & Whyte, J. (2005). The vegetative and minimallyconscious states: Current knowledge and remaining questions.The Journal of Head Trauma Rehabilitation, 20(1), 30–50.

Giacino, J., & Zasler, N. (1997). Outcome after severetraumatic brain injury: Coma, the vegetative state, and theminimally responsive state. The Journal of Head TraumaRehabilitation, 10(1), 40–56.

Giacino, J. T., Ashwal, S., Childs, N., Cranford, R., Jennett, B.,Katz, D. I., et al. (2002). The minimally conscious state:Definition and diagnostic criteria. Neurology, 58(3), 349–353.

Giacino, J. T., & Whyte, J. (2003). Amantadine to improveneurorecovery in traumatic brain injury-associated diffuseaxonal injury: A pilot double-blind randomized trial. TheJournal of Head Trauma Rehabilitation, 18(1), 4–5. authorreply 5–6.

Kotchoubey, B. (2007). Event-related potentials predict theoutcome of the vegetative state. Clinical Neurophysiology,118(3), 477–479.

Laureys, S., Berre, J., & Goldman, S. (2001). Cerebral functionin coma, vegetative state, minimally conscious state, locked-in syndrome and brain death. In J. L. Vincent (Ed.),2001 yearbook of intensive care and emergency medicine(pp. 386–396). Berlin: Springer-Verlag.

Millis, S. R., Rosenthal, M., Novack, T. A., Sherer, M., Nick, T.G., Kreutzer, J. S., Jr., et al. (2001). Long-term neuropsy-chological outcome after traumatic brain injury. The Journalof Head Trauma Rehabilitation, 16(4), 343–355.

Owen, A. M., Coleman, M. R., Boly, M., Davis, M. H.,Laureys, S., & Pickard, J. D. (2006). Detecting awareness inthe vegetative state. Science, 313(5792), 1402.

Rappaport, M., Hall, K. M., Hopkins, K., Belleza, T., & Cope,D. N. (1982). Disability rating scale for severe head trauma:Coma to community. Archives of Physical Medicine andRehabilitation, 63(3), 118–123.

Ritchie, P. D., Cameron, P. A., Ugoni, A. M., & Kaye, A. H.(2000). A study of the functional outcome and mortality inelderly patients with head injuries. Journal of ClinicalNeuroscience, 7(4), 301–304.

Schnakers, C., Perrin, F., Schabus, M., Majerus, S., Ledoux, D.,Damas, P., et al. (2008). Voluntary brain processing indisorders of consciousness. Neurology, 71(20), 1614–1620.

The Multi-Society Task Force on PVS. (1994a). Medicalaspects of the persistent vegetative state (1). The NewEngland Journal of Medicine, 330(21), 1499–1508.

The Multi-Society Task Force on PVS. (1994b). Medicalaspects of the persistent vegetative state (2). The NewEngland Journal of Medicine, 330(22), 1572–1579.

Whyte, J. (2007). Treatments to enhance recovery from thevegetative and minimally conscious states: Ethical issuessurrounding efficacy studies. American Journal of PhysicalMedicine and Rehabilitation, 86(2), 86–92.

Whyte, J., Katz, D., Long, D., DiPasquale, M. C., Polansky, M.,Kalmar, K., et al. (2005). Predictors of outcome in prolongedposttraumatic disorders of consciousness and assessment ofmedication effects: A multicenter study. Archives of PhysicalMedicine and Rehabilitation, 86(3), 453–462.

Yohai, V. J. (1987). High breakdown point and high efficiencyrobust estimates for regression. Annals of Statistics, 15,642–656.