A GENERAL REGRESSION NEURAL-NETWORK ANALYSIS OF PROGNOSTIC MARKERS INPROSTATE-CANCER

This paper assesses the value of general regression neural networks in the analysis of clinical and experimental prognostic factors and in t he prediction of response to treatment and outcome in prostate cancer. 38 patients are considered in this study. The investigation includes a number of established and experimental factors with 3 clinical outco mes: (a) no response to initial treatment, (b) disease relapse and pro gression, and (c) sustained complete response to treatment. An overall classification rate of 89.5% is achieved together with equally high s ensitivity and specificity rates. The results obtained by means of the neural approach offer a significant improvement over those derived th rough classical univariate and multivariate statistical analyses. (C) 1998 Elsevier Science B.V. All rights reserved.