Current issues of ACP Journal Club are published in Annals of Internal Medicine


Clinical Prediction Guide

5 simple clinical variables predicted 1-year survival after traumatic brain injury

ACP J Club. 1999 Nov-Dec;131:81. doi:10.7326/ACPJC-1999-131-3-081


Source Citation

Signorini DF, Andrews PJ, Jones PA, Wardlaw JM, Miller JD. Predicting survival using simple clinical variables: a case study in traumatic brain injury. J Neurol Neurosurg Psychiatry. 1999 Jan;66:20-5.


Abstract

Question

Can a model including variables that are readily available in routine practice predict survival in patients with moderate-to-severe traumatic brain injury?

Design

The model was derived and validated in 2 separate patient cohorts.

Setting

A regional trauma center in Edinburgh, Scotland, United Kingdom.

Patients

The model was derived by using data from 372 consecutive patients (mean age 42 y, 78% men) with traumatic brain injury who were admitted between January 1989 and July 1991 and validated by using data from 520 patients who were admitted between July 1991 and April 1996. Inclusion criteria for both groups were age ≥ 14 years, and an admission or last known Glasgow coma scale (GCS) score ≤ 12 or a GCS score of 13 to 15 plus concomitant systemic injuries that had an Injury Severity Score (ISS) ≥ 16. Follow-up for the derivation sample was 98%.

Description of prediction guide

The model included 5 variables: age, GCS score, ISS, pupillary reaction (1 or neither pupil reactive), and presence of hematoma on computed tomography. By using a simple nomogram, points were assigned for each variable and then summed; the total score was matched to a corresponding probability for survival at 1 year.

Main outcome measure

Survival at 1 year.

Main results

Results of the multivariate logistic analysis for the derivation sample are summarized in the Table. For this sample, the error rate for the predictive model was 10.1% and the area under the receiver-operating characteristic curve was 0.901. For the validation sample, the error rate was 15.2% and the area under the curve was 0.835.

Conclusion

A prediction model that included 5 clinical variables (age, Glasgow Coma Scale score, Injury Severity Score, pupillary reaction, and presence of hematoma on computed tomography of the head) predicted 1-year survival in patients with moderate-to-severe traumatic brain injury.

Source of funding: Medical Research Council.

For correspondence: Dr. D.F. Signorini, Biostatistics Department, Quintiles Scotland Ltd, Inchwood, Bathgate, West Lothian EH48 2EH, Scotland, UK. FAX 44-150-681-3230.


Table. Odds ratios for survival at 1 year after moderate-to-severe traumatic brain injury (derivation cohort)

Variable Odds ratio (95% CI)
No reactive pupils 0.168 (0.06 to 0.50)
Age > 50 y 0.545 (0.43 to 0.69) for each 5-y increase in age
1 reactive pupil 0.599 (0.18 to 2.00)
Injury Severity Score 0.737 (0.60 to 0.91) for each 5-unit increase
Glasgow Coma Scale score 1.31 (1.12 to 1.53) for each unit increase
No visible hematoma on computed tomography 3.53 (1.43 to 8.73)

Commentary

The ability to diagnose a disease and predict its outcome has been a hallmark of a good healer since the beginning of time. Almost 30 years ago, Leaper and colleagues (1) showed that mathematical models could provide a diagnostic accuracy similar to that of a seasoned clinician.

The study by Signorini and colleagues provides further evidence that simple "paper-and-pencil" logistic regression models, using readily available clinical variables, can accurately predict survival in patients with traumatic brain injury. External validation of the prediction method on a subsequent patient sample in the same institution showed slightly lower accuracy. Because the predictor has not been validated outside the original context, its generalizability remains uncertain.

What use could be made of such a predictor? Murray and colleagues (2) have shown how explicit prognosis can affect care. In today's cost-conscious climate, administrators might be tempted to influence clinical management decisions on the basis of an explicit prognosis that can be easily calculated by a clerk. Because the present predictor is excessively pessimistic at low survival probabilities, indiscriminate use could cause a bad prognosis to fulfill itself. It may be prudent to recommend this and similar prediction instruments primarily for stratification in prospective intervention trials rather than for actual clinical decision making.

For functional outcome after traumatic head injury, accurate early prognostic instruments do not exist, largely because functional recovery depends on many factors other than early clinical data.

Ralph Bloch, MD
University of BerneBerne, Switzerland


References

1. Leaper DJ, Horrocks JC, Staniland JR, De Dombal FT. Computer-assisted diagnosis of abdominal pain using "estimates" provided by clinicians. Br Med J. 1972;4:350-4.

2. Murray LS, Teasdale GM, Murray GD, et al. Does prediction of outcome alter patient management? Lancet. 1993;341:1487-91.