Genes retained in multivariate models to predict time from diagnosis to treatment in the training dataset.

Abbreviations: M, model; AIC, Akaike Information Criterion; BIC, Bayes Information Criterion; SAM, sample collection; CM, clinical model.
*Four different sets of genes were chosen as starting points, containing 5, 6, 12, or 13 genes.
†We applied stepwise forward-backward methods to each gene set to optimize either AIC or BIC.
‡Two gene sets optimized predictions of the time from sample collection to treatment; all others looked at the time from diagnosis to treatment.
§Log rank P value to test, via Cox proportional hazards, if a continuous score derived to predict time-to-treatment also predicts overall survival.
¶P-value computed from a chi-squared test of whether the continuous score adds value to the existing clinical predictors.
(c) Individual genes that remained significant when added to the existing clinical predictors.