[A method for the predictive estimation of the surgical risk in adult cardiac pathology]

In order to test the efficiency of statistical predictive models, we compare the results of a standard method (Parsonnet) with the model created through the data of our population.

We used the chi 2 univariate model, lineal and logistic regression with the data of the whole population receiving cardiac surgical procedure from January 1, 1990 to December 31, 1993 (total 1626 patients). The population was divided into a control group (1100 cases, 68%) and a study group (526 cases, 32%). The coefficients of the control group were used to estimate the results in the study group.

Univariate model p value. Significant (p < 0.001) for emergency, age, pulmonary hypertension, left ventricular failure, preoperative use of intra-aortic balloon pump; p < 0.05 mitral valve disease, aortic aneurysm and reoperation. No significance (p < 0.01) was found for gender, aortic or tricuspid disease, percutaneous transluminal coronary angioplasty, unstable or postinfarction angina, transplant, left main or vessel disease number, and mitral, tricuspid or aortic procedure. MULTIVARIATE MODEL: Emergency, pulmonary hypertension, age, left ventricular dysfunction and aortic aneurysm. We estimated a 5.2%, 5.2% and 11.4% mortality with linear, logistic and Parsonnet method respectively with a real group mortality of 6.5%. The average error of the observed and predicted mortality after risk stratification was 5.7%, 6% and 12%.

A model for risk prediction based on the data of the own institution is more accurate for that population than a model created for comparison between institutions, because the former takes account of the center and population peculiarities.