Predicting trauma inpatient mortality in an administrative database: an investigation of survival risk ratios using...

A precise and unbiased measure of injury severity, calculable from administrative data sets, is needed for population-based studies.

The 2001 Nationwide Inpatient Sample was used to calculate independent and traditional survival risk ratios. Logistic regression models predicting survival for New York State's Statewide Planning and Research Cooperative System 2003 inpatients (117,630 records) were constructed for both types of survival risk ratios. The contribution of only the worst survival risk ratio (SRR), the two worst SRRs, the three worst, the four worst, and all trauma diagnoses was investigated. The contribution of gender and age to the models was assessed.

Use of the two worst SRRs, gender, and age gives the best prediction of inhospital mortality.

Inhospital mortality in both trauma centers and in community hospitals can be predicted well from an administrative database by using three predictors--gender, age, and the product of the SRRs for the two diagnoses with the lowest chance of survival.