Classification of observational data with artificial neural networks versus discriminant analysis in pharmacoepidemiological studies - Can outcome offluoxetine treatment be predicted?

For several years, there has been an ongoing discussion about appropriate m ethodological tools to be applied to observational data in pharmacoepidemio logical studies. It is now suggested by our research group that artificial neural networks (ANN) might be advantageous in some cases for classificatio n purposes when compared with discriminant analysis. This is due to their i nherent capability to detect complex linear and nonlinear functions in mult ivariate data sets, the possibility of including data on different scales i n the same model, as well as their relative resistence to "noisy" input. In this paper, a short introduction is given to the basics of neural networks and possible applications. For demonstration, a comparison between artific ial neural networks and discriminant analysis was performed on a multivaria te data set, consisting of observational data of 19738 patients treated wit h fluoxetine. It was tested, which of the two statistical tools outperforms the two other in regard to the therapeutic response prediction from the cl inical input data. Essentially, it was found that neither discriminant anal ysis nor ANN are able to predict the clinical outcome on the basis of the e mployed clinical variables. Applying ANN, we were able to rule out the poss ibility of undetected supressor effects to a greater extent than would have been possible by the exclusive application of discriminant analysis.