Validation of an outcome prediction model for critically ill trauma patients without head injury

J Trauma. 1997 Dec;43(6):934-8; discussion 938-9. doi: 10.1097/00005373-199712000-00011.


Background: The Acute Physiology and Chronic Health Evaluation (APACHE) II system is inaccurate in predicting the risk of death in trauma patients, especially those without head injury. Using multivariate analysis of the APACHE II system in a development set, a new predictive equation was modeled. The four variables that were independently associated with mortality were PaO2/FiO2 ratio, mean arterial pressure, temperature, and the need for inotropic support. This model was tested prospectively in an independent validation set of 300 patients.

Methods: Risk of death was calculated using the APACHE II system with the diagnostic category of multiple trauma and weighting for operative intervention as required. The new model was similarly assessed using the four predictor variables and their beta-coefficients for each mechanism of injury and the entire group. The predicted risk of death derived by both models was compared with the observed mortality rate. Discrimination was calculated using a 2 x 2 decision matrix with a decision threshold of r = 0.5 and receiver operating characteristic curves. Calibration was assessed graphically and by statistical correlation.

Results: The observed mortality rate was 28.3% and the predicted mortality risk was 27.4% for the model and 6.26% for APACHE II. The sensitivity and specificity of the model were 58.8 and 90.7%, and the sensitivity and specificity of APACHE II were 1.2 and 100%. The areas under the receiver operating characteristic curves were 0.84 and 0.78 for the model and the APACHE II system, respectively. Calibration of the model was superior within all deciles of risk (model, R2 = 0.93, p < 0.001; APACHE II, R2 = 0.82, p = 0.02).

Conclusion: The model accurately predicted the risk of death for the entire group. It is superior to the APACHE II system and is the highest reported sensitivity for 24-hour intensive care unit predictive models that have been applied to the critically injured.

MeSH terms

  • Adolescent
  • Adult
  • Aged
  • Critical Illness
  • Discriminant Analysis
  • Female
  • Humans
  • Injury Severity Score
  • Logistic Models*
  • Male
  • Middle Aged
  • Multiple Trauma / classification*
  • Multiple Trauma / mortality*
  • Multivariate Analysis*
  • Outcome Assessment, Health Care
  • Prognosis
  • Prospective Studies
  • Reproducibility of Results
  • Risk Factors
  • Sensitivity and Specificity
  • South Africa
  • Survival Analysis