For each model, a probability of 0.5 or more was defined as predictive of death based on examination of receiver operator characteristic curves. The first model used age and the variables which by univariate analysis were statistically associated (p<.05) with mortality at the time of intubation and the second used these variables and those recorded 24 hours later. Adjusted odds ratios were calculated from the coefficients. Goodness-of-fit of the model was tested by the method of Hosmer and Lemeshow. Interaction among variables was evaluated by the method of Rothman. The included variables in each model were also tested for interaction by the BMDP logistic regression package, and no significant interactions were detected.
For validation, the models were applied to 46 patients admitted to the intensive care unit (ICU) between March 1, 1988, and February 1, 1989. Inclusion criteria and data collection methods were the same as those applied to the original population.
APACHE II scores were assigned as described by Knaus et al,” with the exception that the 24-hour period for data collection began at patient entry into the Emergency Department. Survival was defined as survival to discharge from hospital. Information on mortality following hospital discharge was obtained from contacting patients’ physicians, and from the Monroe Count) Department of Vital Statistics.