jury and outcome at 1 year was analyzed in 170 patients. Seventytwo of ... when the initial GCS scores were in the intermediate range of 5, 6, or 7. No patients ...
J Neurosurg 54:300303, 1981
Early prediction of outcome in headinjured patients BYRON YOUNG, M.D., ROBERT P. RAPP, PHARM.D., J. A. NORTON, R.N., DENNIS HAACK, PH.D., PHILLIP A. TIBBS, M.D., AND JAMES R. BEAN, M.D.
Department of Surgery, Division of Neurosurgery, Department of Pharmacy and the Tobacco and Health Research Institute, University of Kentucky, Lexington, Kentucky The relationship between Glasgow Coma Scale (GCS) scores obtained during the 1st week after head in jury and outcome at 1 year was analyzed in 170 patients. Seventytwo of 76 patients with initial GCS scores of higher than 7 had favorable outcomes. Only two of the 21 patients with initial GCS scores of 3 or 4 lived, and only one had a favorable outcome. Favorable and unfavorable outcomes were almost equally divided when the initial GCS scores were in the intermediate range of 5, 6, or 7. No patients with an initial GCS score in this intermediate range that subsequently worsened had a favorable outcome, while over 80% of those improving to a score higher than 7 had a favorable outcome. Only 12% of those persisting with a score of 5, 6, or 7 for I week had a favorable outcome. Outcome predictions using the multiple logistic model were made for this intermediate group of patients based on GCS scores and data on midline shift derived from computerized tomography (CT). The patients with initial scores of 5, 6, or 7 with midline shifts of less than 4.1 mm on initial CT scanning had a significantly higher favorable outcome rate compared with patients with a larger shift. However, outcome predictions made by combining shift data and initial GCS scores are not significantly more accurate than predictions based solely on initial GCS scores. Combining 48hour GCS scores and shift data significantly improves predictive accuracy based only on coma scores. The data obtained by combining GCS scores at 72 hours and 1 week and shift data is marginally significant for improving accuracy of outcome predictions. It is con cluded that GCS scores and shift data are highly accurate indicators of outcome in headinjured patients.
KEY WORDS
�9
head injury �9 prognosis �9 coma scale �9 computerized tomography
T
HE Glasgow C o m a Scale (GCS) 11 is used in num erous neurosurgical units to evaluate headin jured patients. This report is a detailed statisti cal analysis of the relationship between G C S scores obtained during the 1st week following head injury and outcome at 1 year. The value of additional data obtained from computerized t o m o g r a p h y (CT) scan ning for improving accuracy of outcome predictions based on G C S scores is also assessed. The p a p e r de scribes how, when, and with what accuracy outcome predictions can be made for the individual head injured patient.
Clinical Material and Methods This study consisted of 170 prospectively selected patients with the following types of injuries: pene trating missile wounds, intracranial h e m a t o m a s , depressed skull fractures, and blunt head injuries caus 300
ing major focal neurological deficits or unconscious ness for at least 6 hours. The 6hour duration of un consciousness was selected to provide comparability with other c o m a studies? There were 144 males and 26 females. The patients' average age was 26.2 years, with a range between 2 and 84 years. Although the severity of brain injury varied, all patients required either surgical t r e a t m e n t or m a n a g e m e n t in an inten sive care unit. The definition of coma used in this study conformed to the recommendations of the H e a d Injuries C o m m i t t e e of the World Federation of Neuro surgical Societies, and is currently used in several large collaborative studies. 1,3,Sa,s Patients in c o m a spoke no recognizable words, did not obey c o m m a n d s , and did not open their eyes in response to any stimu lus. Patients with initial G C S scores not accurately measuring the severity of brain injury were removed from the study. For example, patients who greatly improved after correction of anoxia or hypotension,
J. Neurosurg. / Volume 54 / March, 1981
Early prediction of outcome in head injury TABLE 1 Outcome at 1 year in 170 headinjured patients
Initial* GCS Score
No. of Cases
815 76 57 73 34 21 total 170 *GCS = Glasgow Coma Scale.
Favorable Outcome Good Moderate Recovery Disability 63 9 28 8 1
92
0
17
or whose depressed level of consciousness was shown later to be due to alcohol intoxication, were excluded. N o patient was admitted to the study more than 12 hours after injury, and 90% were evaluated within 6 hours of their injury. The G C S was used to grade each patient's level of consciousness. 11 The initial GCS score was obtained immediately after admission and before treatment at our tertiary hospital, although some patients were given osmotic diuretics and dexa methasone by the p r i m a r y physician prior to referral to our unit. Additional G C S scores selected for use in this study were: the highest score in the first 24 hours, and the scores recorded at 48 hours, 72 hours, and 1 week after injury. Clinical m a n a g e m e n t with surgery, intraventricular pressure monitoring, dexametha sone, hypertonic solutions, and barbiturates was indi cated by the patient's brain injury based on current standards of optimal neurosurgical treatment. Each patient was followed for 1 year. Outcome was cate gorized according to the Glasgow Outcome Scale. 6 The outcome categories were subdivided into favor able (good recovery and m o d e r a t e disability) or un favorable outcome (severe disability, vegetative state, and dead). The admission C T scans obtained for pa tients with initial c o m a scores of 5, 6, or 7 were ex amined for evidence of midline shift. Scans were ob tained on 69 of these 73 patients. Measurements for midline shift were m a d e by using the septum pellu cidum and outer tables of the skull as markers. The shift measurements were divided into two groups: less than 4.1 m m or 4.1 m m or greater. A l m m shift measured on our scanner represents a 4.lram actual shift. A l m m m e a s u r e m e n t was the smallest incre ment at which two of the authors (B.Y. and P.A.T.) consistently agreed with the others' independent mea surements. For data analysis and for making out come predictions, we used the multiple logistic model with estimates and standard errors being (iterated) m a x i m u m likelihood. ~ Results
Based on distinct difference in outcome at 1 year, the 170 patients were divided for detailed analysis into three groups (Table 1): initial G C S score greater than 7, initial G C S score of 5, 6, or 7, and initial G C S score J. Neurosurg. / Volume 54 / March, 1981
Unfavorable Outcome Severe Disability 3 9
Vegetative State 0 4
12
5
0
1
Dead 1 24 19 44
of 3 or 4. All but one of the 76 patients with an initial score greater than 7 lived, and only three had an un favorable outcome. Only two of the 21 patients with an initial score of 3 or 4 lived. One m a d e a good re covery after rapid removal of an acute epidural hema t o m a , and the other remained in a vegetative state. Only 36 of the 73 patients with an initial G C S score of 5, 6, or 7 had a favorable outcome. Twentyfour of the 37 with an unfavorable outcome died (Table 1). Accurate outcome predictions cannot be made for the intermediate group with initial scores of 5, 6, or 7, because favorable and unfavorable outcomes occurred in almost equal proportions (Table 1). W e therefore assessed whether adding midline shift data to the G C S score data improves accuracy of o u t c o m e prediction since midline shift, when considered alone, is a signifi cant determinant of outcome when the patients are divided into two groups, shift less than 4.1 m m or 4.1 m m or greater, and outcome categories are combined into favorable or unfavorable groups. The patients with midline shifts of less than 4.1 m m on initial CT scan had a significantly higher favorable outcome rate c o m p a r e d with patients with a larger shift (Table 2). However, predictions made by combining shift data and initial G C S scores are not significantly m o r e ac curate than outcome predictions based solely on ini tial c o m a scores (coefficient 4 SE: 0.60 4 0.58). Age, considered alone, was marginally significant for predicting outcome in this intermediate group (p = 0.09). The average age for the 36 patients with a favorable outcome was 22 years, and 28 years for the 37 patients with an unfavorable outcome. In addition, age category, when subdivided into those less than 15
TABLE 2 Relationship between midline shift and outcome*
Midline Shift Outcome (ram) Unfavorable Favorable 4.1 *X 2 (1) = 5.5; p =0.019.
17 17
27 8
30'1
B. Young, et al. years of age, those aged 15 to 59 years, and those aged 60 years and older, added to initial G C S data did not improve accuracy of outcome predictions (coefficient • SE: 0.15 + 0.61). M o r e accurate predictions, using G C S scores alone, can be m a d e if the score later changes to greater than 7 or less than 5 (Table 3). N o patient who subse quently worsened had a favorable outcome, while over 80% of those improving to a c o m a score higher than 7 had a favorable outcome. Only 21% of those with a G C S score of 5, 6, or 7 that persisted for 1 week had a favorable outcome.
Although adding data on midline shift to initial GCS score data does not increase accuracy of out come predictions, at later intervals midline shift does become an i m p o r t a n t variable for improving outcome predictions (Tables 4 and 5), Combining 48hour G C S scores and shift data significantly improves predictive accuracy. The data obtained by combining G C S scores at 72 hours and 1 week and shift data are mar ginally significant for predicting outcome. Patients persisting with a G C S score of 5, 6, or 7 or improving to a score higher than 7 by 48 hours and having less than 4.1 m m midline shift, fare considerably better than those patients with a shift greater than 4.1 ram.
TABLE 3
Discussion
Percentage o f different time intervals o f patients with initial coma scores o f 5 to 7 and favorable outcomes by I year*
Percent of Cases at 24 Hours 48 Hours 72 Hours 1 Week improved 88 (93) 84 (88) 83 (86) 83 (86) remained same 20 (21) 17 (19) 9 (10) 12 (12) worsened 0 (0) 0 (0) 0 (0) 0 (0) *Percent with favorable outcome after the seven patients with penetrating missile wounds are excluded is shown in parentheses. Outcome
TABLE 4 Value o f adding GCS data to shift data to increase accuracy of outcome predictions at various time intervals
Time Intervals 24 48 72 1 Hours Hours Hours Week improve 3.79 4.40 4.99 4.78 ment scale* • 0.85 • 1.13 • 1.24 • 1.25 shift 1.15 2.34 1.64 1.62 +0.90 • 1.14 +0.93 • *Improvement scale: Glasgow Coma Scale (GCS) score > 7, remains at 5, 6, or 7, or falls below 5. Data
TABLE 5 Probability o f f avorable outcome using GCS score and midline shift data at various time intervals*
Outcome
Midline Shift
24 Hours 94% 82% 25% 10%
improved no (GCS score > 7) yes stayed same no (GCS score yes 5, 6, or 7) worsened no 1% (GCS score < 5 ) yes 0% *GCS = Glasgow Coma Scale. 302
Time Interval 48 72 1 Hours Hours Week 96% 92% 92% 68% 69% 69% 21% 7% 9% 3% 1% 2% 0% 0%
0% 0%
0% 0%
The Glasgow C o m a Scale is of value not only as an objective and reproducible means of quantifying the degree of neurological impairment, but also as a basis for making early, accurate predictions of the likely outcome of headinjured patients. Initial G C S scores alone provide adequate information to m a k e highly reliable outcome predictions for patients with initial scores either higher than 7 or less than 5. Ninetyfive percent of patients with initial G C S scores greater than 7 had a favorable outcome, and 95% with initial scores of less than 5 had an unfavorable outcome. Correspondence between outcome and G C S scores at later time intervals does not improve for these two groups during the remainder of the 1st week. Jennett, et al., related G C S scores to outcome in several reports, and indicated that assessment of severity of injury " o n admission or before resuscita tion and early t r e a t m e n t " will bias outcome predic tions. 7 In Jennett's view, comparable injuries could be mistakenly estimated to have a less favorable o u t c o m e if initial or worst state rather than "best state" is the basis for m a k i n g the prediction. He indicated that using the best state observed during the first 24 hours more reliably predicts outcome. Our data show that the initial level of consciousness can be used to reliably predict the o u t c o m e in patients with an initial G C S score higher than 7 or less than 4. Waiting 24 hours to determine highest G C S score in the first 24 hours for this group of patients would not increase predictive ac curacy when anoxia, hypotension, or alcohol intox ication are not contributing to the alteration of consciousness. We found that assessing prognosis for patients in the intermediate group with G C S scores of 5, 6, or 7 was more complex. Initial scores did not provide suf ficient information to m a k e accurate predictions since outcomes were divided almost equally between the f a v o r a b l e and u n f a v o r a b l e categories. A c c u r a t e predictions for this group can be made only when the GCS score later changes to greater than 7 or less than 5. For example, when the initial G C S score improved to at least 8 during the first 24 hours, 88% had a favor able outcome. When deterioration to below 5 oc curred, no patient survived. The earlier the patient im J. Neurosurg. / Volume 54 / March, 1981
Early prediction of outcome in head injury proves into the higher than 7 category, the more likely is a favorable outcome. The chances of having a favorable outcome reach a plateau at 24 hours for those patients with initial GCS scores of 5, 6, or 7 who improve to scores of greater than 7 (Table 3). The longer the patient remains in the intermediate category, the more likely is an unfavorable outcome. Patients remaining in the intermediate category (5, 6, or 7) at 1 week had only about onehalf the chance of having a favorable outcome of those in this same category at 24 hours. N o patient who was initially in the intermediate group and then deteriorated had a favorable outcome. Our data support Jennett's find ing that the pattern of change in GCS score during the first 24 hours after injury is a key to more precise early outcome predictions within this group. Even more ac curate predictions can be made at later intervals, but the favorable outcome percentages reach a plateau at the highest score in the first 24 hours. Stablein, et al., 1~ recently detailed the statistical value of the logistic model for prediction outcome of headinjured patients. In their series, the most impor tant prognostic factor was a requirement for surgical decompression. In a previous description of their stan dardized protocol for managing these patients, Becker, et a l . ? indicated that patients haVing a 5mm or more shift had surgical exploration and decom pression. Although Stablein, et al., and Becker, et al., did not include penetrating missile wounds, our studies are comparable. Table 3 shows no significant difference in favorable outcome percentages after seven patients with penetrating missile wounds are ex cluded. Our data also emphasize the significant re lationship between shift in the 45 mm range and outcome. In addition, we show the value of com bining shift data with G C S score data for improving accuracy of outcome predictions for patients in the in termediate category. Other factors, such as pupillary reactivity to light, brainstem function, vital signs and hematocrit, age, and type o f injury can be used as a basis to predict out come following head injury. 2.9 Our data suggest that adding any one or a combination of these other factors could only very slightly increase the already highly ac curate outcome predictions of patients who are ini tially conscious or deeply comatose or who are in the intermediate group at 24 hours after admission. Our study shows that age, considered independently in the intermediate group, is a marginally significant vari able for predicting outcome. However, after adjust ment for initial coma score, age is not a significant variable for this group. Likewise, injury category when added to initial GCS score does not improve ac curacy of outcome predictions (coefficient 4 SE: 0.31 + 0.21). Stablein, et al.,1~ based their outcome predictions on 12 variables obtained initially after hospital admis sion. Ninetyone percent of their outcome predictions were correct. Sixtynine percent of their predictions were made at a 0.90 confidence level. They showed J. Neurosurg. / Volume 5 4 / March, 1981
that requirement for a surgical decompression (shift greater than 5 mm), age, vital signs and hematocrit, and m o t o r response (a component of the G C S scale) are significant determinants of outcome. The other eight factors they considered, including pupillary light responses, pupillary size, and oculocephalic response, were either only marginally significant or not signifi cant. We found that only one Variable, the initial G C S score, is needed for an accurate prediction of outcome in patients deeply comatose or conscious. Accurate predictions using only the GCS score cannot be made a few hours after admission for patients with initial scores of 5, 6, or 7. However, by 24 hours, accurate outcome predictions based only on G C S scores are possible. Furthermore, adding only one variable, mid line shift data, will significantly improve the outcome predictions made at later intervals. Our study shows that a short delay enables highly accurate predictions for the intermediate group without using so many variables that a computer is required for accurate predictions. References
1. Bates D, Caronna J J, Cartlidge NEF, et al: A prospec tive study of nontraumatic coma: methods and results in 310 patients. Ann Neurol 2:211220, 1977 2. Becker DP, Miller JD, Ward JD, et al: The outcome from severe head injury with early diagnosis and inten sive management. J Neurosurg 47:491502, 1977 3. Bruce DA, Schut L, Bruno LA, et al: Outcome follow ing severe head injuries in children. J Neurosurg 48: 679688, 1978 4. Cox DR: Analysis of Binary Data. London: Methuen and Co, 1970 5. Frowein RA: Classification of coma. Acta Neurochir 34:510, 1976 6. Jennett B, Bond M: Assessment of outcome after severe brain damage. A practical scale. Lancet 1:480484, 1975 7. Jennett B, Teasdale G, Braakman R, et al: Prognosis of patients with severe head injury. Neurosurgery 4: 283289, 1979 8. Jennett B, Teasdale G, Galbraith S, et al: Severe head injuries in three countries, d Neurol Neurosurg Psy chiatry 40:291298, 1977 9. Langfitt TW: Measuring the outcome from head in juries. J Neurosurg 48:673678, 1978 10. Stablein DM, Miller JD, Choi SC, et al: Statistical methods for determining prognosis in severe head in jury. Neurosurgery 6:243248, 1980 II. Teasdale G, Jennett 13: Assessment of coma and im paired consciousness. A practical scale. Lancet 2:8183, 1974 This work was supported by NINCDS Grant NS13454 03. This paper was presented at the Annual Meeting of the American Association of Neurological Surgeons, April 2024, 1980, New York, New York. Address reprint requests to: Byron Young, M.D., Univer sity of Kentucky Medical Center, MS 107, 800 Rose Street, Lexington, Kentucky 40536. 303