Computer-derived equations for predicting survival postoperatively. Their usefulness and limitations.
We used multivariate analysis to determine whether survival following perforations of the gastrointestinal tract could be accurately predicted from preoperative data. Of 12 variables tested, four were found to have predictive value. These were age, pulmonary disease, preoperative shock, and the attending surgeon. When these four variables were employed in a logistic regression equation on 42 patients, it correctly predicted which 21 patients died before leaving the hospital. To produce an equation useful for other hospitals, we recalculated it without the attending surgeon variable. Again, the equation was used to predict survival. The correlation of predicted vs observed outcome remained high, and, using a 2 x 2 chi 2 test, the correlation was significant. We then cross validated the three-variable model on data from a second hospital. The model accurately predicted the new data equally well. We believe that predictive models can identify risk factors in a variety of patient populations and can determine who is likely to benefit from specific treatment modalities.