Improved Disease Outcome Prediction Based on Microarray and Clinical Data Combination and Pre-validation

boosting, clinical data, combined models, generalized linear models, logistic regression, microarray data, model evaluation

Combining relevant information from high-dimensional microarray data and low-dimensional clinical variables to predict disease outcome is important to improve treatment decisions. Such a combination may yield more accurate predictions than those obtained based on the use of microarray or clinical data alone. We propose a combination of logistic regression for clinical data and BinomialBoosting for microarray data. Then we propose its extension designed for redundant sets of data. Our approach combines microarray and clinical data at the level of decision integration. The extension includes pre-validation of models built with microarray and clinical data followed by weights calculation. Weights determine relevance of microarray and clinical models for data combination. Evaluations are performed with several redundant and non-redundant simulated datasets. Then some tests are applied to two real benchmark datasets. Our approach increases outcome prediction on non-redundant simulated datasets and does not decrease outcome prediction on redundant simulated datasets. Pre-validation of built models improves outcome of the prediction up to 4% in the case of real redundant dataset.