Risk stratification simplified: the worst injury predicts mortality for the injured children

J Trauma. 2008 Dec;65(6):1258-61; discussion 1261-3. doi: 10.1097/TA.0b013e31818cac29.


Background: The International Classification Injury Severity Score (ICISS) uses anatomic injury diagnoses to predict probability of survival (Ps) computed as the product of the survival risk ratios (SRR) of the three most severe injuries. SRRs are derived as the proportion of fatalities for every International Classification of Diseases-9th Revision-Clinical Modification diagnosis in a "benchmark" population. Pediatric-specific SRRs were computed from 103,434 entries in the National Pediatric Trauma Registry. We hypothesized that ICISS was a valid pediatric outcome predictor, and that the child's most severe injury; i.e., the lowest SRR, is the major driver of outcome, which can be used alone to predict survival.

Methods: Receiver operator characteristic analysis was used to assess the predictive validity of ICISS. SRRs derived from 53,235 phase II patients were used as the training set to calculate the Ps for 50,199 phase III children comprising the test set. The survival probability (Ps) computed from the standard three diagnoses was compared with that computed from only the worst injury (lowest SRR). Records with a single diagnosis or Ps of 1, indicating no mortality potential, were excluded from the analysis. Nagelkerke pseudo R2 defined what proportion of the predicted Ps was the effect of the worst injury alone versus the traditional Ps.

Results: A total of 25,239 records with at least two diagnoses with SRRs indicating risk of mortality were analyzed. The area under the receiver operator characteristic curve for traditional Ps was 0.935, compared with 0.932 for that calculated using only the lowest SRR. The difference of 0.003 was not significant (z = 1.061, p = 0.2888, NS). Nagelkerke pseudo R2 for the lowest SRR was 0.455 compared with 0.462 for the traditional three diagnosis Ps, which shows that the majority of Ps predictive power is related to the single injury with the lowest SRR. Further analysis demonstrated that this effect was related to frequency of coexistent injuries with no mortality risk rather than definable difference in severity.

Conclusion: These data validate ICISS as predictive of pediatric injury survival. The dominant effect of the worst injury reflects an epidemiologic characteristic of pediatric trauma that will identify specific injuries for best practice analysis and focused injury prevention.

Publication types

  • Comparative Study

MeSH terms

  • Child
  • Hospital Mortality
  • Humans
  • Multiple Trauma / classification
  • Multiple Trauma / diagnosis
  • Multiple Trauma / mortality*
  • Outcome Assessment, Health Care / statistics & numerical data
  • Probability
  • ROC Curve
  • Registries
  • Reproducibility of Results
  • Retrospective Studies
  • Risk Assessment
  • Survival Analysis
  • Trauma Severity Indices*
  • Wounds and Injuries / classification
  • Wounds and Injuries / diagnosis
  • Wounds and Injuries / mortality*