a prognostic model for head injury
TRANSCRIPT
Acta Neurochirurgica 45, 199--208 (1979) ACTA NEUROCHIRURGICA �9 by Springer-Verlag 1979
State University of New York Upstate Medical Center Syracuse, New York, U.S.A.
A Prognostic Model for Head Injury
By
W. A. S tewar t , S. P. Li t ten, and P. R. Sheehe
With 2 Figures
S u m m a r y
A prognostic model for head injured patients was developed. Patients fall into one of four prognostic categories at the end of the first hospital day:
I. Discharge alive on or before the seventh hospital day. II. Discharge alive 8-42 days after admission. III. Discharge alive after 42 days, or dead after five years. IV. Discharge dead before five years. The outcome of 630/0 of the patients was predicted correctly. The model cor-
rectly placed, or missed by only one category, 97~ of the patients observed. Errors tended to the optimistic side of observed outcomes. Age and pattern of consciousness were critical prognostic factors.
I n t r o d u c t i o n
A classification (Table 1) thought to increase the accuracy of early prediction of head injury outcome was proposed in 1973 ~ Though all conceivable situations are not described in this categoriz- ation, it accounted for essentially all injuries in our experience. The basis for this classification was the pattern of the post-traumatic level of consciousness. Initial testing of this classification was done by developing a prognostic model for severe diffuse brain injuries (BI), a subgroup of the classification. Patients so defined met criteria for homogeneity as they moved through different levels of care in their post<raumatic course to endpoints of death or resumption of pre-traumatic activity. Age, presence or absence of skull fracture, and superimposed mass lesion were the most useful prognostic factors of the large number of variables tested.
13"
0001-6268/79/0045/0199/$ 02.00
o
Tab
le
1. H
ead
Inju
ry
Cla
ssifi
catio
n by
Pos
t-Tr
aum
atic
P
atte
rn o
f C
onsc
ious
ness
Typ
e of
inj
ury
Pat
tern
of
cons
ciou
snes
s M
otor
R
efle
xes
Pup
ils
Res
pira
tion
Fat
al (
F)
vari
able
; co
mat
ose
and
unre
spon
sive
fl
acci
d ab
sent
di
late
d an
d ap
neic
or
on e
xam
inat
ion
unre
acti
ve
inad
equa
te
Seve
re d
iffu
se b
rain
co
ma
pers
istin
g fo
r da
ys t
o m
onth
s sp
onta
neou
s an
d/or
sm
all
and
ofte
n in
tuba
tion
in
jury
(B
I)
wit
h pr
olon
ged
reco
very
per
iod
refl
ex d
ecer
ebra
te o
r un
reac
tive
re
quir
ed
deco
rtic
ate
post
urin
g;
mus
cle
tone
inc
reas
ed
DT
R:
incr
ease
d su
perf
icia
l:
abse
nt
brai
n st
em :
depr
esse
d as
ymm
etri
c
Luc
id i
nter
val
(LI)
co
ma;
im
prov
emen
t; d
eter
iora
tion
va
riab
le
vari
able
va
riab
le
vari
able
Pr
ogre
ssiv
e in
jury
wit
hout
com
a fo
llow
ed
vari
able
va
riab
le
vari
able
va
riab
le
dete
rior
atio
n (P
D)
by d
eter
iora
tion
L
imite
d co
ntus
ion
(LC
) co
ma
follo
wed
by
retu
rn t
o no
rmal
va
riab
le
usua
lly
inta
ct
vari
able
in
tuba
tion
ov
er a
per
iod
of m
inut
es t
o ho
urs
not
requ
ired
an
d ra
rely
wee
ks
A Prognostic Model for Head Injury 201
Other investigators have attempted to quantify the prognostic reliability of age, coma, and other observed variables in an effort to develop formal classifications or models that will predict the end result of a head injury.
Kerr, Kay, and Lassman 4 examined the outcomes of civilian adults who lost consciousness through head injury. These authors found that a level of consciousness gradient gave the highest re- gression coefficient for predicting death in hospital. Mortality in- creased with age, and surgical shock was of death-predicting im- portance at all ages, especially 50 years and older. Knowledge of sex, skull fracture status, and intracranial haemorrhage added little prognostic information.
Carlsson, yon Essen, and Lofgren 1 examined a series of severely brain injured patients defined as those who died in the first 12 hours of hospitalization or who were comatose during the period. Older people surviving primary unconsciousness had increased risk of developing complications leading to death. A formula for predicting time from trauma until restitution was proposed: restitution time (days) = coma duration (days) 5< 15 + 1.5 (age in years). Another formula gave restitution propabilities which depended on age and duration of coma. The dependence of these formulas upon the highly variable duration of coma limits their usefulness. Each outcome (death and restitution) was considered separately in the analysis, and failure to reach either outcome was neglected.
Jennett e t al . 3 attempted to predict head injury outcome in a study of 600 patients treated for severe craniocerebral trauma at hospitals in the Netherlands and Great Britain. Predictions of out- come at six months were made in 200 randomly selected patients using clinical examination features thought to have predictive power. Outcomes from the 200 patients were compared with those from the remaining 400 cases. Death or survival was correctly predicted at the 95~ probability level in 44~ of the cases after 24 hours of observation, and in 61~ after collecting data for three days. In- clusion of patients with severe injuries to other organ systems com- promises the reliability of the study.
This study compared the prognostic ability of a model based on age alone with one which was also based on the history of pattern of consciousness as reported at the time of admission. The latter was then compared with a model which also utilized the observed pattern of consciousness during the first 24 hours of hospitalization. Analysis of the data resulted in a method placing patients with head injury into one of four prognostic groups after the first 24 hours of hospitalization. These prognostic groups are:
202 W.A. Stewart et al.:
I. Discharge alive on or before the seventh hospital day. II. Discharge alive 8-42 days after admission. III. Discharge alive after 42 days, or dead after five years. IV. Discharge dead before five years.
Patient Population
The medical records of all patients admitted to the Upstate Medical Centre Emergency Room between 1. July 1965 and 31. December 1972 with a diagnosis of head injury were examined. Included in the review were all charts having diagnos- tic class numbers ~ corresponding to: cerebral concussion, cerebral contusion, traumatic subarachnoid heamorrhage, subdural or extradural haemorrhage, other intracranial haemorrhage, and intracranial injury. Patients with simple concussion were excluded from the study. Also omitted were I0 patients with substantial associated trauma such as multiple long bone fractures, extensive soft tissue injury, crushed chest, pulmonary injury, spine injury, or multiple organ trauma. Three hundred and fifty two patients provided the material for this study. Patients with regional soft tissue injury, single extremity fracture, and facial injuries that did not compromise the airway were accepted into the study. Their distribution by source of injury and age is shown in Table 2.
Table 2. Number (Per Cent) of Patients by Source of Trauma for Each Age Group
Trauma Age Totals
0-14 15-24 25-54 --> 55
Fall 32 (49) 12 (10) 15 (15) 35 (54) 94 (27) Vehicular 26 (40) 85 (70) 70 (69) 13 (20) 194 (55) Assault 2 (3) 12 (10) 4 (4) 7 (11) 25 (7) Other 5 (8) 12 (10) 12 (12) 10 (15) 39 (11)
Totals 65 (100) 121 (100) 101 (100) 65 (100) 352 (100)
Prognostic Factors
The original classification (Table 1) was enlarged by addition of a sixth group (UNK) to accommodate patients for whom information regarding the level of consciousness from time of injury to initial examination was unavailable or contra- dictory. This unknown group contained 150 patients. On analysis these patients behaved like those with limited contusion (LC) injuries.
Age was represented by four categories: 0-14 years; 15--24 years; 25-54 years; and 55 years and older.
The model using pattern of consciousness as observed during the first 24 hours of admission categorized patients as follows: comatose on admission and 24 hours later (U-U); conscious on admission and 24 hours later (C-C); comatose on admis- sion and conscious 24 hours later (U-C); and conscious on admission and comatose 24 hours later (C-U).
Coma or unconsciousness was defined as "loss of physiological and motor re-
A Prognosuc Model for Head Injury 203
sponse to stimulation, or reduction of these responses to rudimentary motor re- flexes"5. In other words these patients lay flat with eyes closed, aphonic, and unresponsive to verbal stimuli. Spontaneous motor activity, when present, was not purposeful. Response to tactile or noxious stimuli, when present, was reflex in nature. Any other neurological status was considered to be "conscious" or "awake" for the purpose of this study.
Methods
An outcome index combining morbidity (duration of hospitaliz- ation) and mortality (dead or alive at discharge) was constructed; e.g., discharged alive after 1 day ranked higher than discharged alive after nine months, which was higher than discharged dead after six months, which was higher than discharged dead during the first hospital day. The prognostic categories described above were derived from this index. A model based on the original classification (Table 1) and age was compared with the age alone model using multiple regression techniques. Further comparisons were made by introducing variables from the model based on pattern of conscious- ness as observed in the hospital. Each observed outcome was com- pared to that indicated by multiple regression. Each patient was given a misclassification weight of 0, 1, 4, or 9, increasing as the square of the discrepancy between the observed and predicted out- come group. The sum of misclassification weights was used to score the performance of each model. A search for the best model (lowest misclassification score) was made using the multiple regression equa- tion and various reductions of it.
Results
Evaluation of Prognostic Factors
The misclassification score for the regression model based on age and history of post-traumatic pattern of consciousness was 248, very significantly smaller than the score of 372 for the best regression model based on age alone (P < 0.005) (Table 3).
The inclusion of variables identifying the pattern of consciousness during the first 24 hours of hospital admission resulted in further substantial reduction of the misclassification score from 248 to 157. A simplified version of this model was obtained, without apparent loss in prognostic capacity, by excluding interactive variables, merging the three youngest age classes, and merging BI, LI, and PD into one group. The outcome ranks predicted by this model, presented in detail below, are converted into predicted outcome groups ac-
204 W . A . Stewart et al.:
Table 3. Prediction of Outcome; Age and History of Pattern o~ Consciousness *
Predicted Diagnostic outcome class-age groups categories
Observed outcome categories
I I I I I I IV
Totals
LC 0-14 15
LC 15-24 29
LC 25-54 27
10 0 0 " 25
9 2 1 41
14 1 0 42
Subtotals 71 33 3 1 108
II LC ~>55 5 3 0 1 9
U N K 0-14 9 7 0 1 17
U N K 15-24 33 21 5 1 60
U N K 25-54 24 7 5 2 38
U N K -> 55 10 14 5 6 35
PD 0-14 7 5 2 1 15
PD 15-24 1 2 1 0 4
LI 15-24 1 1 2 0 4
Subtotals 90 60 20 12 182
I I I PD 25-54 0 2 1 1 4
PD -->55 0 4 3 3 10
LI 0-14 0 2 2 0 4
LI 25-54 2 2 0 1 5
LI ~<55 0 2 i 4 7
B1 15-24 0 1 7 2 10
Subtotals 2 13 14 11 40
IV BI 0-14 0 0 0 2 2
BI 25-54 0 0 4 6 10
BI ~>55 0 0 0 3 3
F 0-14 0 0 0 2 2
F 15-24 0 0 0 2 2
F 25-54 0 0 0 2 2
F >--55 0 0 0 1 1
Subtotals 0 0 4 18 22
Totals 163 106 41 42 352
* Misclassification score -- 248.
A Prognostic Model for Head Injury 205
cording to an auxiliary table, also presented below *. This procedure correctly discriminated the outcome of 63~ of the patients (Fig. 1). More specifically, outcome was correctly predicted in 80, 52, 46, and 38~ of the patients for categories I to IV, respectively. The model correctly placed, or missed by only one category, 97~ of the patients observed (shaded area in Fig. 1). Prediction errors tended to be on the optimistic side of the observed outcome.
Predictec
71-
TOt
T~r
Totals I 165 106
Fig. 1. The best model correctly placed 63~ of the patients (circled). Ninety-seven per cent of the patients were placed correctly, or missed by only one category
(shaded area)
For practical convenience, the multiple regression equation and the auxiliary table are summarized by Fig. 2, which may be used like a nomogram to assign patients to one of the four outcome categories previously described.
* The multiple regression equation for the predicted outcome rank (Y) of a ^
patient is: Y = 1 2 9 . 4 + 3 6 . 9 X ~ + 3 0 . 1 X 2 + 6 5 . 4 X 3 - - 2 0 . 5 X 4 + 145.1Xs. The five regressors are defined as follows: X1 = 1 if the patient is ~> 55 years of age; otherwise Xi = 0; X2 = 1 if the history is compatible with LI, PD, BI injury; otherwise X2 = 0; X3 = 1 if the history is compatible with fatal injury; other- wise X~ = 0; X4 ~ I if the history and 24 hours observation permits classification as LC/CC; otherwise X4 = 0; X5 = 1 if the patient is unconscious at 24 hours; otherwise X5 = 0. The following table is used to convert the predicted outcome
A
rank (Y) to a predicted outcome group:
Predicted outcome group
~< 157 I 158-270 II 271-360 III
361 IV
206 W.A. Stewart eg al.:
D i s c u s s i o n
The ability to predict the outcome of a head injury early in the post-traumatic period is helpful for several reasons. Socio-economic effects of the injury on the patient and family may be evaluated more realistically. Accurate estimates of the cost of head injuries to society in monetary and human resources are made possible. Ef- ficient organization of personnel and facilities in a regional trauma service is expedited. This information encourages responsible use of public funds which support medical programmes for care of trauma
Pattern of Consciousness in Hospital
Admission 24 hrs. Later
AWAKE AWAKE
COMA AWAKE
C O / ~ " / COMA / AWAKE
~N.A. = Not Applicable
Pattern of Consciousness Age by H is tory
LC UKN LI, PD, BI Fatal
, 5 5 N A "
< 5 5 ~ / / z / /_../A Z / / /
Fig. 2. Shows the prognostic factors used to assign patients to the ourcome cate- gories of the best model
victims. Level of care provided to patients may be adjusted to the predicted outcome. New types of treatment may be evaluated more objectively.
An ealier study 6 and the findings of this report confirm age and pattern of post-traumatic consciousness as useful variables for predict- ing the outcome of head injury. Age is ideally suited for use in prognostic models. The pattern of post-traumatic consciousness is more difficult to quantify. The careful grading or evaluation of the level of consciousness may provoke differences of opinion even among experts. For this reason gross shades of the level of conscious- ness (unconscious or conscious as defined above) were selected for use on the assumption that among all observers there would be a higher level of agreement.
In this study patient outcomes ranged from discharge alive after one day to institutionalization lingering over five years, to death on the first day. A simple method of ranking patients in a unified morbidity and mortality index that encompassed this variety of
A Prognostic Model ~or Head Injury 207
outcomes was developed. The four outcome groups reported in this study were derived from this outcome index.
Of the four prognostic groups described in this study, categories one and four are the most useful. Patients who fall into category one will usually return to pre-traumatic activity. Category four identifies patients whose injury is fatal. The crucial questions of survival or death, and quality of survival, are answered for patients falling into these categories for practical purposes.
The model is least reliable when it predicts that a patient will fall into category three. In none of the first three categories does this model forecast the quality of life which the patient may expect following recovery from the injury.
Age was not as critical a prognostic factor for this wide spectrum of head injuries, heavily weighted by mild trauma, as it was in the study of severe diffuse injuries 6
The shortcomings of the model are most aparent in categories 1 and 2. These attest to the difficulty of formalizing the prognosis of outcome from serious head injury. Further efforts in this direction should consider prolonging observation of the pattern of post- traumatic consciousness. More precise distinctions of levels of con- sciousness other than a division between wakefulness and coma may improve prognostic efforts. Attempts to identify milestones in the post-traumatic course that permit refocussing of the initial prognosis are warranted. Such an approach proved to be useful in prognosis for patients with severe diffuse brain injuries. While updating of prognoses might occur at arbitrary times, it seems more advantageous to be able to do so as the patient reaches a specific level of improve- ment or deterioration.
Prospective testing of this model in a new population of head injuries and its use by other groups of neurosurgeons to predict out- come in their patients may be done to confirm or refute the findings of this study.
C o n c l u s i o n s
Morbidity and mortality data for patients with head injuries were merged to form a single ordered set of outcome categories. While age alone was useful in predicting outcomes, significant im- provement was achieved by also using the pattern of post-traumatic consciousness up to the time of hospitalization as well as during the first 24 hours after admission.
A table (Fig. 2) converts age and consciousness data into cor- responding prognostic categories.
208 W.A. Stewart et aI.: A Prognostic Model for Head Injury
R e f e r e n c e s
1. CarLson, C. A., yon Essen, C., Lofgren, J., Factors affecting the clinical course of patients with severe head injuries. J. Neurosurg. 29 (1968), 242--251.
2. Commission on professional and hospital activities, hospital-international classification of diseases, adapted. Ann Arbor, 1968.
3. Jennett, B., Teasdale, G., Braakman, R., Minderhoad, V., Knill-Jones, R., Predicting outcome in individual patients after severe head injury. Lancet i (1976), 1031--1034.
4. Kerr, T.A., Kay, D. W. K., Lassman, L.P., Characteristics of patients, type of accident and mortality in consecutive series of head injuries admitted to a neurosurgical unit. Brit. J. Prey. Soc. Med. 25 (1971), 179--185.
5. Plum, F., Posner, J. B., The diagnosis of stupor and coma, p. 2. Philadelphia: F. A. Davis Company. 1966.
6. Stewart, W.A., Litten, S.P., Sheehe, P.R., A prognostic model for brain stem injury. Surg. Neurol. 1 (1973), 303--310.
Authors' addresses: W. A. Stewart, M.D., 725 Irving Avenue, Syracuse, NY 13210: S. P. Litten, M.S., State University of New York College of Forestry, Syracuse, NY 13210; P. R. Sheehe, Sc.D., State University of New York Upstate Medical Center, Syracuse, NY 13210, U.S.A.