Computer-aided decision support systems enable physicians to make more accurate clinical decisions and can significantly improve the quality of care provided to patients. However, prediction of classification confidence as the degree of reliability on the resulting predictions is a much needed step in clinical decision making. A recently developed technique called conformal prediction utilizes the similarity between a new sample and the training samples in order to form confidence measures for predictions. However, the conventional conformal prediction method suffers from shortcomings such as high computational complexity that prevent its use in real-time applications. This paper introduces an alternative approach to the conventional confidence prediction that addresses some of this and other disadvantages. Both real clinical and non-clinical datasets are employed to test and validate the capabilities of the proposed approach.