Fig. 4: Performance of the classification model discriminating cancer from non-cancer in patients presenting with non-specific symptoms of cancer.

A Overview of the classification model pipeline. B Venn-diagram showing the results of univariate feature selection in the training set within the discovery cohort. P-values were obtained using two-sided t-test and adjusted for multiple testing using the Benjamini–Hochberg method. C Key features selected by the lasso model. D ROC curves in the discovery and replication cohorts. E Distribution of model probabilities for cases and controls in the discovery and replication cohorts. The dashed lines represent the median cancer probability.