This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
O
PTJCX
HDa1
f(a
j
d
mcLeaatdr
wefCr
CicsCpsFS
DH
sz
aaR
1495
RIGINAL ARTICLE
redictors of Extended Rehabilitation Length of Stay Afterraumatic Brain Injury
uan Carlos Arango-Lasprilla, PhD, Jessica M. Ketchum, PhD, David Cifu, MD, Flora Hammond, MD,amilo Castillo, MD, Elizabeth Nicholls, MT, Thomas Watanabe, MD, Anthony Lequerica, PhD,
iaoyan Deng, MS
h
M
TmUt$mLcwhbl
cbetidilmbswatmr
ftrTm
ABSTRACT. Arango-Lasprilla JC, Ketchum JM, Cifu D,ammond F, Castillo C, Nicholls E, Watanabe T, Lequerica A,eng X. Predictors of extended rehabilitation length of stay
fter traumatic brain injury. Arch Phys Med Rehabil 2010;91:495-504.
Objective: To develop a prediction rule for acutely identi-ying patients at risk for extended rehabilitation length of stayLOS) after traumatic brain injury (TBI) by using demographicnd injury characteristics.
Design: Retrospective cohort study.Setting: Traumatic Brain Injury Model Systems.Participants: Sample of TBI survivors (N�7284) with in-
uries occurring between 1999 and 2009.Interventions: Not applicable.Main Outcome Measures: Extended rehabilitation LOS
efined as 67 days or longer.Results: A multivariable model was built containing FIMotor and cognitive scores at admission, preinjury level of edu-
ation, cause of injury, punctate/petechial hemorrhage, acute-careOS, and primary payor source. The model had good calibration,xcellent discrimination (area under the receiver operating char-cteristic curve � .875), and validated well. Based on this model,formula for determining the probability of extended rehabilita-
ion LOS and a prediction rule that classifies patients with pre-icted probabilities greater than 4.9% as at risk for extendedehabilitation LOS were developed.
Conclusions: The current predictor model for TBI survivorsho require extended inpatient rehabilitation may allow for
nhanced rehabilitation team planning, improved patient andamily education, and better use of health care resources.ross-validation of this model with other TBI populations is
ecommended.
From the Departments of Physical Medicine and Rehabilitation (Arango-Lasprilla,ifu, Castillo, Nicholls), Biostatistics (Ketchum, Deng), and the Center for Rehabil-
tation Sciences and Engineering (Arango-Lasprilla, Ketchum, Cifu, Castillo, Ni-holls, Deng), Virginia Commonwealth University, Richmond, VA; Indiana Univer-ity School of Medicine, Indianapolis, IN (Hammond); Carolinas Rehabilitation,harlotte, NC (Hammond); Moss Rehab/Albert Einstein Medical Center, Philadel-hia, PA (Watanabe); Department of Physical Medicine and Rehabilitation, Univer-ity of Medicine and Dentistry of New Jersey, Newark (Lequerica); the Kessleroundation, West Orange, NJ (Lequerica); and Physical Medicine and Rehabilitationervice, Richmond Veterans Administration Medical Center, Richmond, VA (Cifu).Supported by the National Institute on Disability and Rehabilitation Research, U.S.
epartment of Education (grant nos. H133A070036, H133A21943-16, and133A07003).No commercial party having a direct financial interest in the results of the research
upporting this article has or will confer a benefit on the authors or on any organi-ation with which the authors are associated.
Reprint requests to Juan Carlos Arango-Lasprilla, PhD, Dept of Physical Medicinend Rehabilitation, Virginia Commonwealth University, Rehabilitation Psychologynd Neuropsychology, VCU West Hospital, 3rd Fl, Rm 3-102, 1200 E Broad St,ichmond, VA 23298, e-mail: [email protected].
RAUMATIC BRAIN INJURY is a major cause of mor-bidity and mortality in the United States. Of the esti-
ated 1.5 to 2.0 million persons who sustain TBI in thenited States each year, nearly 235,000 will require hospi-
alization, resulting in long-term economic consequences of4.5 billion.1 TBI often requires significant and expensiveedical and surgical interventions, including extendedOSs during acute care and inpatient rehabilitation, whichreates significant short-term economic consequences asell.2 A culture of pervasive health care cost consciousnessas resulted in decreasing overall LOSs in acute and reha-ilitation settings during the past decade, a trend that isikely to continue.3
Despite a growing emphasis on standardization in TBIare, treatment of patients with this condition continues toe highly variable in nature and cost.4,5 The influence ofarly (acute) and sociodemographic variables on rehabilita-ion LOS and inpatient rehabilitation costs have been stud-ed. Several factors have been identified as significant pre-ictors of rehabilitation LOS,5 including rehabilitationntensity,6 abnormal findings from computed tomography,ength of acute hospitalization, FIM score at admission,5,7
edical complications,7-10 age,5,9,11 presence of intracranialleeds, skull fractures, length of acute hospitalization,5 andeverity of injury.5,9 Specifically, longer rehabilitation LOSsere seen in patients with lower functional independence at
dmission,5,7 such medical complications as extremity frac-ures and/or respiratory problems,7-10 older age,5,9,11 and
ore severe TBIs.5,9 Increased rehabilitation intensity waselated to shorter rehabilitation LOS.6
To our knowledge, no studies have attempted to identifyactors that influence patients who specifically have ex-ended rehabilitation LOSs or attempted to build a multiva-iable model for predicting extended rehabilitation LOS.herefore, the goal of the present study was to build aultivariable model for predicting extended rehabilitation
List of Abbreviations
AUC area under the curveDAI diffuse axonal injuryGCS Glasgow Coma ScaleGED general equivalency diplomaLOS length of stayMVC motor vehicle collisionTBI traumatic brain injury
1496 REHABILITATION LENGTH OF STAY AFTER BRAIN INJURIES, Arango-Lasprilla
A
OS for TBI survivors by using demographic and injuryospital characteristics. By using this model, we aim toreate a multivariable prediction rule that could be usedarly in the rehabilitation process to identify patients at riskor extended rehabilitation LOS. This information may helpealth care providers and family members in planning andnticipating issues that might need to be addressed early toinimize the impact of these factors on rehabilitation LOS.
n addition, this study might provide findings relevant foruture research of the effect on rehabilitation LOS whenhese risk factors are modified.
METHODS
articipantsParticipants were enrolled in the national database of the
ational Institute on Disability and Rehabilitation Re-earch–funded TBIMS program, a multicenter longitudinaltudy of TBI outcomes. Each TBIMS-funded center re-eived approval by its individual institutional review board.nclusion in the TBIMS national database requires the pa-ient to meet the following criteria: (1) at least 16 years oldt the time of injury; (2) arrived at a TBIMS acute-careospital within 72 hours of injury; (3) received acute care,mmediately followed by inpatient rehabilitation within theospitals designated by the TBIMS site; and (4) gave in-ormed consent. According to the TBIMS, TBI is defined asrauma to brain tissue caused by an external mechanicalorce, evidenced by loss of consciousness, posttraumaticmnesia, skull fracture, or objective neurologic findings thateasonably can be attributed to TBI on a physical or mentaltatus examination.
A sample of TBI survivors (N�7526) with injuries oc-urring between January 1, 1999, and January 1, 2009, waselected from the TBIMS database. The dates were selectedo obtain the most recent decade of complete data. Partici-ants had to have a recorded rehabilitation LOS to bencluded in the study. There were 242 participants (3.22%)rom the initial sample of 7526 who could not be included inhis analysis because information about rehabilitation LOSas missing. Participants with missing rehabilitation LOSata were enrolled predominantly in 2003 and 2004 whenhe TBIMS temporarily did not collect data for dates ofhort-term interruptions in rehabilitation. Thus, data from284 participants were analyzed.A random sample of 1821 participants (25%) was held back
rom the definition, preliminary analysis, and primary analysesmodel-building) stages to be used for validating the predictionodel, leaving a sample of 5463 participants for model build-
ng.
easuresDependent variable. The dependent variable of interest
s rehabilitation LOS. This is a continuous variable measur-ng the total number of days in rehabilitation care. Short-erm interruptions (�30d) were subtracted from the mea-ures as appropriate. Because data were not normallyistributed, rehabilitation LOS measures for each participantn the model-building sample initially were log transformed.he transformed data followed a normal distribution reason-bly well. Mean � SD log rehabilitation LOS was estimatedo be 3.01�.724. Based on this, extended rehabilitation LOSas defined as log rehabilitation LOS at or beyond 1.645 SD
f the log mean (ie, log LOS�3.01�1.645�.724) to identify C
rch Phys Med Rehabil Vol 91, October 2010
he expected top 5% of the sample. This definition corre-ponds to a back-transformed rehabilitation LOS of 67 daysr longer (ie, rehabilitation LOS�exp[3.01�1.645�.724]).pproximately 5.1% of the model-building sample and.8% of the validation sample were classified as having anxtended rehabilitation LOS.Independent variables. Demographic measures available
or analysis included age at injury, race/ethnicity, level ofducation, and employment status at injury. Age was mea-ured in years and modeled as a continuous variable. Race/thnicity was self-reported and categorized as white, black,ispanic, or other (including Asian/Pacific Islander, Nativemerican, and unclassified). Level of education was re-
oded into 3 groups: less than high school (�11th grade),igh school/GED (high school diploma, GED, or trade), orore than high school (some college or an associate, bach-
lor, master, or doctoral degree). Employment status wasategorized as competitively employed, unemployed, orther (student, special education, homemaker, retired, vol-nteer, other).Injury and hospital characteristics available for analysis
ncluded GCS score at emergency admission, motor andognitive FIM scores at rehabilitation admission, extent ofntracranial compression (midline shift), presence of punc-uate/petechial hemorrhage, presence of cranial complica-ions (intracranial hypertension), and LOS in acute care.ause of injury was categorized as fall, vehicular/sports,iolent, or other (including falling/flying objects and pedes-rian injuries). Injury severity was determined from the GCScore at admission to the emergency department and usedhe standard classification of severe (score, 3– 8), moderatescore, 9 –11), or mild (score, 13–15). The FIM is an 18-itemeasure of independence rated from admission to rehabili-
ation, with scores for each item, ranging from 1 (requiringull assistance) to 7 (complete independence). The separateotor (13 items) and cognitive (5 items) domains wereodeled as continuous variables.12,13 Extent of intracranial
ompression was categorized as none; cisterns present withidline shift of 1 to 5mm; cisterns compressed or absent,
ut midline shift of 0 to 5mm; or midline shift greater thanmm. Punctate/petechial hemorrhage and cranial complica-ions (intracranial hypertension) were coded as present orbsent. LOS in acute care was measured in days.
In addition, information was available pertaining to aatient’s history of excessive alcohol use, primary personiving with the participant at the time of admission, primaryesidence at injury, and primary rehabilitation payor source.xcessive alcohol use was determined based on the numberf occasions the participant had 5 or more drinks during theonth before injury: none, 1 to 4, or 5 or more occasions.rimary person living with the participant at injury waslassified as alone, family member (spouse, parent, sibling,hild, significant other, other relative), or other (roommate,ther patients, other residents, personal care attendant,ther). Primary residence at injury was categorized as pri-ate residence, medical facility (nursing home, adult home,cute/rehabilitation hospital, subacute care), or other (cor-ectional institution, hotel/motel, homeless, other). Primaryehabilitation payor source was categorized as commercialworkers’ compensation, Blue Cross and Blue Shield, otherrivate insurance, health management organization, no-faultnsurance, preferred provider organization, Civilian Healthnd Medical Program of the Uniformed Services), Medicaidtraditionally or managed care administered, pending, state
1497REHABILITATION LENGTH OF STAY AFTER BRAIN INJURIES, Arango-Lasprilla
lly or managed care administered), or uninsured (privateay, hospital [free bed], other).
tatistical AnalysisAll statistical analyses were conducted by using SAS,
ersion 9.2.a Demographic and injury characteristics wereescribed by using mean � SD values for continuous vari-bles and counts and percentages for categorical variables.o assess for potential bias, the sample of 7284 participants
hat was not missing rehabilitation LOS data (eligible) wasompared with the sample of 242 participants missing reha-ilitation LOS data (ineligible) with respect to demographicnd injury characteristics by using t tests for continuousariables and chi-square tests for categorical variables.The following steps were taken by using the model-
uilding sample to build a multivariable model for predict-ng extended rehabilitation LOS: (1) significant univariateredictors of extended rehabilitation LOS were identified bysing simple logistic regression models (P�.05), (2) a mul-ivariable logistic regression model was fit containing allignificant univariate predictors, and (3) the multivariableodel was reduced by removing insignificant predictor vari-
bles (P�.05) 1 at a time. Calibration and discriminationere assessed by using the Hosmer-Lemeshow goodness-f-fit test and the AUC for both the model-building andalidation samples. A threshold in the predicted probabilityas chosen to maximize sensitivity while maintaining at
east 75% specificity.
rediction RuleWe aimed to create a prediction rule that would be simple
o use and provide the user with robust estimates of theredicted probability of extended rehabilitation LOS andisk classification. All decisions about model specificationsere made by using the model-building sample. After val-
dation, an updated formula for the predicted probability andpdated prediction rule was determined based on logisticegression coefficients estimated by using the entire eligibleample (combining the model-building and validation sam-les). Calibration and discrimination were reassessed bysing the updated model.
RESULTS
escription of the SampleDemographic and injury characteristics of the sample of
articipants included in the model-building and validationtages are listed in table 1. For the entire sample, averagege of participants was 39.6�18.7y, and participants wererimarily men (73%), white (70%), employed at injury64%), and had at least a high school level of education atnjury (71%). Injuries were predominately vehicular/sportelated (58%) and classified as moderate or severe (60%).hose classified as having an extended rehabilitation LOSad a median rehabilitation LOS of 88 days (interquartileange, 76 –109), whereas those classified as not having anxtended rehabilitation LOS had a median rehabilitationOS of 19 days (interquartile range, 12–29).
ligible Versus Ineligible ParticipantsDemographic and injury characteristics of the ineligible
ample of participants are listed in table 1 (right columns).
neligible participants had lower FIM motor (30.6 vs 35.6; p
�.001) and cognitive (14.6 vs 16.0; P�.009) admissioncores and lower percentages of punctate/petechial hemor-hages (17% vs 26%; P�.001) compared with eligible par-icipants. In addition, ineligible participants had lower per-entages of vehicular/sports-related injuries (52% vs 58%)nd greater percentages of other related injuries (14% vs%) compared with eligible participants (P�.029). No otherariables were significantly different between groups (all�.05).
nivariate Predictors of Extended Rehabilitation LOSDistributions of demographic and injury characteristics
re listed for patients in the model-building sample with andithout extended rehabilitation LOSs in table 2. Age at
njury, level of education, admission GCS score, cause ofnjury, admission motor and cognitive FIM scores, extent ofntracranial compression, punctate/petechial hemorrhage,ranial complications (intracranial hypertension), acute-careOS, and primary payor source had significant univariate
elationships with extended rehabilitation LOS. In general,igher percentages of extended rehabilitation LOS weressociated with younger age, higher levels of education atnjury, more severe admission GCS scores, violent and otherelated injuries, lower admission motor or cognitive FIMcores, any midline shift, presence of punctate/petechialemorrhage, presence of cranial complications (intracranialypertension), longer stays in acute care, and commercialnsurance or Medicaid as the primary payor source. Thereere no significant univariate relationships between ex-
ended rehabilitation LOS and sex, race/ethnicity, employ-ent status at injury, excessive alcohol use, primary person
iving with participant at injury, and residence at injury.ecause admission GCS score and cranial complications
intracranial hypertension) had high rates of missing data�20%), these variables were not included in multivariablenalyses.
ultivariable Prediction of Extended Rehabilitation LOSA multivariable logistic regression model was built to
redict extended rehabilitation LOS and initially includedll significant univariate predictors previously identified.ge at injury (P�.677) and extent of intracranial compres-
ion (P�.470) did not contribute significantly to the modelfter adjusting for the other predictor variables and subse-uently were removed. Parameter estimates for the finalodel are listed in table 3 (left columns). Reference cate-
ories were chosen arbitrarily as less than high school,njuries caused by falls, no punctate/petechial hemorrhage,nd uninsured payor sources. Adjusted odds ratios for eachredictor variable are listed in table 4 (left columns). Similaro what was found with univariate analyses, extended reha-ilitation LOS was associated with lower FIM motor andognitive scores, longer stays in acute care, higher levels ofducation, injuries related to other causes (vs falls, vehicu-ar/sports, or violence), presence of punctate/petechial hem-rrhage, and use of commercial insurance or Medicaid (vsedicare or being uninsured).The final model based on the model-building sample had
ood calibration (Hosmer-Lemeshow goodness-of-fit test,�.963) and excellent discrimination (AUC�.878). The
lotted in fig 1. A threshold (cutoff value) in the predicted
Arch Phys Med Rehabil Vol 91, October 2010
1498 REHABILITATION LENGTH OF STAY AFTER BRAIN INJURIES, Arango-Lasprilla
A
Table 1: Demographic and Injury Characteristics of Model-Building, Validation, and Ineligible Samples
Independent Variable
Model Building(n�5463)
Validation(n�1821)
Ineligible(n�242)
Mean�SD Mean�SD Mean�SD
Age at injury (y) 39.54�18.76 39.84�18.66 41.38�20.10Admission motor FIM score 35.71�19.31 35.28�18.88 30.64�17.87Admission cognitive FIM score 15.95�7.89 16.04�7.97 14.60�7.88Acute LOS (d) 20.48�15.96 20.47�17.40 21.61�15.76
Excessive alcohol use (�5 drinks)None 3163 67.98 1024 66.54 156 75.001–4 times in mo before injury 800 17.19 266 17.28 30 14.42�5 times in mo before injury 690 14.83 249 16.18 22 10.58
Primary person living with at injuryAlone 907 16.63 315 17.35 49 20.25Family 4083 74.85 1329 73.18 171 70.66Other 465 8.52 172 9.47 22 9.09
1499REHABILITATION LENGTH OF STAY AFTER BRAIN INJURIES, Arango-Lasprilla
Table 2: Demographic and Injury Characteristics by Extended Rehabilitation LOS for the Model-Building Sample
Independent Variable
Not Extended(RLOS �67d)
Extended(RLOS �67d) Comparison
Mean�SD Mean�SD t/df P
Age at injury (y) 39.67�18.85 37.14�16.80 t5461�2.19 .029Admission motor FIM score 36.71�19.14 15.90�9.11 t378.6�32.91* �.001Admission cognitive FIM score 16.34�7.80 8.59�5.58 t5408�16.27 �.001Acute LOS (d) 19.71�15.17 34.87�22.24 t291.0��11.23* �.001
bbreviation: RLOS, rehabilitation LOS.Unequal-variance t test used.
Arch Phys Med Rehabil Vol 91, October 2010
pateg
afi.o
A*
1500 REHABILITATION LENGTH OF STAY AFTER BRAIN INJURIES, Arango-Lasprilla
A
robability of .050 was associated with sensitivity of 86.0%nd specificity of 75.2%. Thus, the prediction rule based onhis model is to classify a person as being at risk forxtended rehabilitation LOS if the predicted probability isreater than 5%.
Table 3: Multivariable Prediction M
Parameter
Model Buildi
Estimate � SE �2
Intercept �2.3474�.4392Admission motor FIM score �0.0736�.0098 �2
1�
Admission cognitive FIM score �0.0678�.0168 �21�
Acute-care LOS (d) 0.0219�.0036 �21�
Level of education �22�
�High school 0High school/GED 0.2870�.1947�High school 0.6606�.1934
Cause of injury �23�
Fall 0Vehicular/sports �0.0292�.2185Violent 0.4310�.2956Other 1.0835�.2732
Admission motor FIM score 0.929 (0.911–0.947)Admission cognitive FIM 0.934 (0.904–0.966)Acute-care LOS (d) 1.022 (1.015–1.029)Level of education
HS/GED vs �HS 1.332 (0.910–1.952)�HS vs �HS 1.936 (1.325–2.828)�HS vs HS/GED 1.453 (1.045–2.020)
Cause of injuryVehicular/sports vs fall 0.971 (0.633–1.490)Violent vs fall 1.539 (0.862–2.747)Other vs fall 2.955 (1.730–5.048)Violent vs vehicular/sports 1.584 (0.985–2.548)Other vs vehicular/sports 3.042 (1.994–4.643)Other vs violent 1.920 (1.083–3.403)
Punctate/petechial hemorrhagePresent vs absent 1.436 (1.068–1.929)
Primary payor sourceCommercial vs Medicaid 0.941 (0.663–1.334)Commercial vs Medicare 1.909 (1.018–3.583)Commercial vs uninsured 2.103 (1.106–3.998)Medicaid vs Medicare 2.030 (1.041–3.956)Medicaid vs uninsured 2.236 (1.146–4.361)Medicare vs uninsured 1.101 (0.464–2.613)
bbreviations: CI, confidence interval; HS, high school; NS, not significaStatistical significance at ��.05.
rch Phys Med Rehabil Vol 91, October 2010
The model applied to the validation sample (N�1821)lso had good calibration (Hosmer-Lemeshow goodness-of-t, P�.746) and excellent discrimination capabilities (AUC�
859) (see fig 1). The prediction rule based on a cutoff valuef .05 yielded sensitivity of 80.9% and specificity of 74.7%.
l for Extended Rehabilitation LOS
Updated
P Estimate � SE �2/df P
�2.6288�.3913�.001 �0.0718�.0083 �2
1�74.46 �.001�.001 �0.0605�.0142 �2
1�18.09 �.001�.001 0.0229�.0029 �2
1�64.85 �.001.002 �2
2�16.54 �.0010
0.2496�.17120.6568�.1697
�.001 �23�23.53 �.001
00.0380�.19120.5285�.25770.8874�.2428
.016 �21�9.83 .002
00.4143�.1321
.021 �23�11.68 .009
00.8493�.29790.8158�.31020.3046�.3888
Odds Ratios
Updated
Significance OR 95% CI Significance
S 0.931 (0.916–0.946) SS 0.941 (0.915–0.968) SS 1.023 (1.017–1.029) S
NS 1.284 (0.918–1.795) NSS 1.929 (1.383–2.690) SS 1.503 (1.127–2.004) S
NS 1.039 (0.714–1.511) NSNS 1.696 (1.024–2.811) SS 2.429 (1.509–3.909) S
NS 1.633 (1.080–2.471) SS 2.338 (1.603–3.412) SS 1.432 (0.864–2.372) NS
S 1.513 (1.168–1.961) S
NS 1.034 (0.761–1.406) NSS 1.724 (1.009–2.945) SS 2.338 (1.304–4.192) SS 1.667 (0.942–2.952) NSS 2.261 (1.231–4.153) S
NS 1.356 (0.633–2.906) NS
ode
ng
/df
56.7416.3737.7912.39
28.71
5.76
9.76
sted
ing
nt; OR, odds ratio; S, significant.
cdmmae(Liowpaidpmtp
petwMioattipapcTc
1501REHABILITATION LENGTH OF STAY AFTER BRAIN INJURIES, Arango-Lasprilla
The model-building and validation samples then wereombined to refit the model by using the same set of pre-ictor variables selected for the final model that used just theodel-building sample. Parameter estimates were reesti-ated by using the larger data set to obtain more accuracy
nd to be used for the updated prediction rule. Parameterstimates based on the updated model are listed in table 3right columns). This model had good calibration (Hosmer-emeshow goodness-of-fit, P�.706) and excellent discrim-
nation (AUC�.875) (see fig 1). An updated threshold (cut-ff value) in the predicted probability of .049 was associatedith sensitivity of 85.1% and specificity of 75.2%. Thus, therediction rule based on the updated model is to classifyperson as being at risk for extended rehabilitation LOS
f the predicted probability is greater than 4.9%. The up-ated prediction rule and the formula for determining theredicted probability of an individual based on the updatedodel are shown in appendix 1. A Web-based clinical
ool for calculations is available at http://go.vcu.edu/erlos-
Fig 1. Receiver operating characteristic (ROC) cu
redictor/. i
DISCUSSIONThe purpose of the present study was to build a model for
redicting extended rehabilitation LOS for TBI survivorsarly after injury by using demographic and injury charac-eristics. Approximately 5.1% of the model-building sampleere defined as having an extended rehabilitation LOS.odel-building strategies using 75% of the sample resulted
n a model containing punctate/petechial hemorrhage, causef injury, FIM motor and cognitive scores at admission,cute-care LOS, primary payor source, and level of educa-ion at injury. The model developed showed good calibra-ion and excellent discrimination by using the model-build-ng, validation, and combined samples. An updatedrediction rule was defined that classified patients as beingt risk for extended rehabilitation LOS if their predictedrobability was greater than 4.9%. This threshold was asso-iated with high sensitivity (85.1%) and specificity (75.2%).his model can be used to predict rehabilitation LOS in thelinical setting for patients by using the formula summarized
1502 REHABILITATION LENGTH OF STAY AFTER BRAIN INJURIES, Arango-Lasprilla
A
The present study expanded on previous research in aumber of areas. Although previous researchers have exam-ned the relationship between neuroimaging findings andehabilitation LOS, they have not specifically focused on aroup of patients with an extended rehabilitation LOS afterBI. Cowen et al5 found that the presence of intracranialemorrhage with skull fracture was associated with a longerehabilitation LOS in general. However, a study of patientsnrolled in the TBIMS database between 1988 and 1995ound that neither the presence of intracranial hemorrhageor the presence of head fractures significantly lengthenedehabilitation LOS.9
In the present study, the presence of punctuate/petechialemorrhage was associated with extended rehabilitationOS in both univariate and multivariate analyses. A radio-raphic finding of this type of hemorrhage is consistent withhe presence of DAI, which is associated with more severend diffuse injury to various structures of the brain.14 There-ore, it is not surprising that extended rehabilitation LOS,lso associated with severity of injury,9,15 would be seen ingroup of patients with DAI. It should be noted that theore common imaging studies (magnetic resonance imaging
nd computed tomography) are not sensitive enough todentify the presence of DAI because the damage often isicroscopic; therefore, the DAI cases identified in the
resent study may represent a group with more severenjuries, further strengthening the relationship between theresence of punctuate/petechial hemorrhage and extendedehabilitation LOS.
In addition to more generally severe injuries, DAI wasssociated with acceleration/deceleration events involving aotational component. DAI often was diagnosed after MVC,
finding also shown in animal models of injury.16 There-ore, it is somewhat surprising that in our multivariateodel, injuries related to causes other than MVC were
ssociated with an extended rehabilitation LOS. One poten-ial explanation is that pedestrian injuries (which would fallnto the “other” category) also can be associated with DAI.n addition, pedestrian injuries often result in both brain andodily injury, which might be related to other medicalomplications that researchers have associated with longerehabilitation LOS.7,9
Acute-care LOS typically was associated with the severity ofither the primary TBI or secondary injuries that accompanyhe TBI (ie, fractures, organ injury). Therefore, it is not sur-rising that the present study found extended rehabilitationOS to be associated with an increasing acute-care LOS. Other
actors that contributed to an increased acute-care LOS, such asospital-acquired infection, medical complications, and limitedunding of health care needs that prevent admission to inpatientehabilitation initially, also would be expected to significantlyncrease rehabilitation LOS. These findings are consistent withhose of High et al,9 who found that initial GCS score, averageuration of impaired consciousness, admission FIM score, andcute-care LOS accounted for 48% of the variance in rehabil-tation LOS. The physical and mental deconditioning that oc-urs with prolonged acute-care hospitalization would be ex-ected to delay and lengthen the overall rehabilitation process.In this study, violence-related TBI was not related to
xtended rehabilitation LOS. This is consistent with findingsf prior studies because previous researchers comparingutcomes for violence-related and nonviolent TBI haveound shorter acute-care and rehabilitation LOSs in theiolence-related groups.17 It is possible that a shorter reha-ilitation LOS in persons with violence-related TBI is re-
ated to fewer associated bodily injuries.18,19 f
rch Phys Med Rehabil Vol 91, October 2010
Lower FIM motor and cognitive scores at admission alsoere associated with extended rehabilitation LOS. Thisnding was consistent with previous studies examining theelationship between FIM score and rehabilitation LOS,5,20
uggesting that patients who arrive at the acute rehabilita-ion setting requiring more assistance may need more timend therapy to reach functional independence. This findingtrongly supports the notion that, as part of a comprehensiveehabilitation assessment of a person with TBI, evaluatinglinicians should perform a complete FIM assessment. Iden-ifying patients who are at high risk for an extended reha-ilitation LOS because of functional impairments is bestone before admission to inpatient rehabilitation so thateam planning, goal planning, development of family expec-ations, notification of the insurer, and other key aspects ofrogramming can be performed as early as possible. Imple-entation of preadmission FIM evaluations may necessitate
hanges in how interdisciplinary team care is provided andequire FIM certification by the acute-care team. Anotherey issue related to this finding is the potential importancef increased acute-care rehabilitation interventions (ie, ther-py) to improve level of functioning. Although future re-earch is needed to show the applicability and efficacy ofhis recommendation, it is supported by the available re-earch on the utility of increased acute-care therapy.5,20
The present model also found primary payor source to beelated to rehabilitation LOS, with patients who had com-ercial insurance or Medicaid more likely to have an ex-
ended rehabilitation LOS. Medicaid funding frequently pre-ents rehabilitation providers with difficulties transitioningatients to the next level of care if additional resources areequired (eg, transfer to a skilled nursing facility or substan-ial assistive devices for transitioning to the home). Com-ercial insurance often applies a long-term perspective of
utcome toward allocating resources to cover the inpatientehabilitation needs of their clients.
In the present study, education was significantly related toxtended rehabilitation LOS, even after controlling forause of injury, punctate/petechial hemorrhage, acute-careOS, and primary payor source. In addition, education re-ained a significant predictor of extended rehabilitationOS when adjusting for employment status. The relevancef this finding is unclear. Future studies should attempt toalidate or verify the relationship between educational at-ainment and rehabilitation LOS.
The implications of these research findings are manifold.irst, these findings have identified key elements of the
njury and acute care for that injury (and related difficulties)hat are associated with prolonged inpatient rehabilitationOS. Interventions to prevent or modify these acute factors,specially the acute-care LOS and level of functioning, mayelp prevent prolonged stays in rehabilitation. Second,hereas specific injury (ie, cause of injury, presence ofunctate/petechial hemorrhages) or patient demographic (ie,evel of education) variables cannot be altered after thenjury, the presence of these specific factors can be identi-ed early after injury. Their presence should alert the reha-ilitation clinician to the increased risk for prolonged inpa-ient care. As noted, changes can be made to the planningrocess for inpatient rehabilitation based on this knowledge,anging from the timing and setting of rehabilitation to theocus and goal setting of the team and family. Finally, earlyse of structured functional assessment (ie, the FIM) whilehe patient is still in the acute-care setting being considered
or inpatient rehabilitation care may allow for a better ability
tlioOutshp
S
l
avtmceoTappfpitatr
t
Na
1
1
1
1
1503REHABILITATION LENGTH OF STAY AFTER BRAIN INJURIES, Arango-Lasprilla
o predict which patients are most likely to require a pro-onged rehabilitation LOS. This data-driven approach tonpatient rehabilitation admissions and planning is preferredver the “wait-and-see” approach most commonly used.verall, results of this study provide a model that can besed by physicians and administrators to identify risk fac-ors for extended rehabilitation LOS. Rehabilitation per-onnel then can use this model to customize individual re-abilitation interventions to target the unique needs ofatients.
tudy LimitationsThese results should be interpreted in light of the following
imitations.1. Caution should be used when generalizing these results
outside the TBIMS setting until they have been validatedin other settings.
2. It also is possible that some variables that were not partof the TBIMS database could affect extended rehabili-tation LOS (ie, current psychological functioning; psy-chological and medical history).
3. Information about medical complications during acutecare and rehabilitation were not analyzed, and thereasons for the extended rehabilitation LOS are notknown. The relevance of the extended rehabilitationLOS to eventual outcome is not known and was notpart of this study.
4. Results of this study only apply to patients who aretransferred from an acute-care facility designated bythe TBIMS directly to acute inpatient rehabilitation.Thus, not included are those patients transferred fromnondesignated acute-care sites and those admittedlater, postinjury, from a late or secondary rehabilita-tion stay.
5. Existing admission criteria and payor source limita-tion may have created a biased sample of patientsadmitted for inpatient care rather than all patients withacute TBI.
CONCLUSIONS
The purpose of the present study was to build a model forcutely predicting extended rehabilitation LOS for TBI sur-ivors by using demographic and injury characteristics ando create an easy-to-use prediction rule. A multivariableodel for prediction was built by using FIM motor and
ognitive scores at admission, acute-care LOS, level ofducation at injury, cause of injury, punctate/petechial hem-rrhage, and primary payor source as predictor variables.he model had good calibration, excellent discrimination,nd validated well. An updated model for computing theredicted probability of extended rehabilitation LOS and arediction rule for identifying at-risk patients were createdor use by health care professionals. The present earlyredictor model for TBI survivors who require extendednpatient rehabilitation may allow for enhanced rehabilita-ion team planning, improved patient and family education,nd better use of health care resources. Cross-validation ofhis model with other populations of patients with TBI isecommended.
Acknowledgments: We thank Brian J. Bush for contributions in
he area of Web-site development.
APPENDIX 1: MULTIVARIABLE PREDICTION RULE FOREXTENDED REHABILITATION (eR) LOS (DEFINED AS >67d)
AFTER TBI
OTE: FIMmotor and FIMcognitive denote FIM scores at rehabilitation, admission,nd education represents the highest level of education attained before injury.
References1. Max W, Mackenzie EJ, Rice DP. Head injuries: cost and conse-
quences. J Head Trauma Rehabil 1991;6:76-91.2. McGarry LJ, Thompson D, Millham FH, et al. Outcomes and cost
of acute treatment of traumatic brain injury. J Trauma 2002;53:1152-9.
3. Kreutzer JS, Kolakowsky-Hayner SA, Ripley D, et al. Chargesand length of stay for acute and inpatient rehabilitation treatmentof traumatic brain injury. Brain Inj 2001;15:763-74.
4. Ottenbacher KJ, Smith PM, Illig SB, Fielder RC, Granger CV.Length of stay and hospital readmission for persons with disabil-ities. Am J Public Health 2000;90:1920-3.
5. Cowen TD, Meythaler JM, DeVivo MJ, et al. Influence of earlyvariable in traumatic brain injury on Functional IndependenceMeasure scores and rehabilitation length of stay and charges. ArchPhys Med Rehabil 1995;76:797-803.
6. Blackerby WF. Intensity of rehabilitation and length of stay. BrainInj 1990;4:167-73.
7. Tooth L, McKenna K, Strong J, et al. Rehabilitation outcomes forbrain injured patients in Australia: functional status, length of stayand discharge destination. Brain Inj 2001;15:613-31.
8. Englander JS, Cifu DX, Wright J, et al. The impact of acutecomplications, fractures, and motor deficits on functional outcomeand length of stay after traumatic brain injury: a multicenteranalysis. J Head Trauma Rehabil 1996;11:15-26.
9. High WM, Hall KM, Rosenthal M, et al. Factors affecting hospitallength of stay and charges following traumatic brain injury.J Head Trauma Rehabil 1996;11:85-96.
0. Lew HL, Lee E, Date ES, Zeiner H. Influence of medical comorbidi-ties and complications on FIM change and length of stay duringinpatient rehabilitation. Am J Phys Med Rehabil 2002;81:830-7.
1. Frankel JE, Marwitz JH, Cifu DX, et al. A follow up study of olderadults with traumatic brain injury: taking into account decreasinglength of stay. Arch Phys Med Rehabil 2006;87:57-62.
2. Linacre JM, Heinemann AW, Wright BD, et al. The structure andstability of the Functional Independence Measure. Arch Phys MedRehabil 1994;75:127-32.
1504 REHABILITATION LENGTH OF STAY AFTER BRAIN INJURIES, Arango-Lasprilla
A
the Functional Independence Measure. Arch Phys Med Rehabil1993;74:566-73.
4. Adams JH, Doyle D, Ford I, et al. Diffuse axonal injury in headinjury: definition, diagnosis, and grading. Histopathology 1989;15:49-59.
5. Lehmkuhl LD, Hall KM, Mann N, Gordon WA. Factors thatinfluence costs and length of stay of persons with traumatic braininjury in acute care and inpatient rehabilitation. J Head TraumaRehabil 1993;8:88-100.
6. Foda MA, Marmarou A. A new model of diffuse brain injury inrats, part II: morphological characterization. J Neurosurg 1994;80:301-13.
7. Harrison-Felix C, Zafonte R, Mann N, et al. Brain injury as a resultof violence: preliminary findings from the traumatic brain injury
model systems. Arch Phys Med Rehabil 1998;79:730-7. a
rch Phys Med Rehabil Vol 91, October 2010
8. Bogner JA, Corrigan JD, Mysiw WJ, Clinchot D, Fugate L. Acomparison of substance abuse and violence in the prediction oflong-term rehabilitation outcomes after traumatic brain injury.Arch Phys Med Rehabil 2001;82:571-7.
9. Esselman PC, Dikmen SS, Bell K, Temkin NR. Access to inpa-tient rehabilitation after violence-related traumatic brain injury.Arch Phys Med Rehabil 2004;85:1445-9.
0. Valach L, Selz B, Signer S. Length of stay in the rehabilitationcenter, the admission Functional Independence Measure and thefunction independence measure gain. Int J Rehabil Res 2004;27:135-43.
Supplier
. SAS Institute, 100 SAS Campus Dr, Cary, NC 27513.