A two-stage predictor strategy is introduced in the context of high dimensional data (large p, small n). Here the focus application is a medical one: prediction of symptomatic infection given molecular expression levels in blood. The first stage of the two-stage predictor uses the previously introduced method of Predictive Correlation Screening (PCS) to select a subset of genes that are important in the prediction of symptom scores. Selected genes are used in the second stage to learn a predictor for the prediction of symptom scores. Under sampling budget constraints we derive the optimal sample allocation rules to the first and second stages of the two-stage predictor. Superiority of the proposed predictor relative to the well known method of LASSO is shown via experiment.