Are decision trees way around some statistic methods?

We are trying to find a replacement for the statistical method of finding significant attributes, using a set of data and a well know method from the field of machine learning-decision trees. The data used for testing contained 90 records, each described by 24 attributes, measured before and after surgery. The surgery was performed on children aged from 2 to 10 years. The purpose of making these measurements was to find those attributes which are significant for determining a predisposition to two forms of acidemia: metabolic acidosis and respiratory acidosis. By applying various statistical methods to the set of records we established which attributes are statistically significant. These results were then compared to results obtained from using the same data set to build different decision trees. We showed that some statistical methods and decision trees can give almost identical results, and the decision trees do not require special knowledge of statistics. The results were also carefully examined by medical experts.