Generation and validation of a primary-tumor-derived four-gene prognostic signature for recurrence (R) of stages I/II colorectal cancer (CRC) following potentially curative resection

Abstract

4035 Background: Current guidance for postoperative clinical management of stages I/II CRC patients (pts) is suboptimal. We hypothesized that a molecular prognostic test using primary CRC tissue would better predict the chances of tumor R within 3 years (y) than NCCN Clinical Practice Guidelines. Methods: Archival formalin-fixed paraffin-embedded (FFPE) primary adenocarcinoma tissues obtained at initial resection with curative intent were retrieved for 145 stage I/II (pT1–4 pN0 M0) CRC pts from multiple international sites; none had neoadjuvant or adjuvant therapy. Pts had tumor R by 36 months (mo) or confirmed non-recurrence (NR) for ≥36 mo after surgery. Pts were stratified by R/NR status and then randomized to a Training Set (TSet) (n=73; 34R, 39NR) or Validation Set (VSet) (n=72; 33R, 39NR). Tumor gene expression was quantified by DASL assay (Illumina, San Diego) using a custom 512-gene panel. Genetic programming (GP), a machine-learning technique, defined 15 TSet genes as key for differentiating pts with R versus NR. Results: Successive GP analyses of TSet data evolved a prognostic signature that mathematically combined 4 of the 15 key genes identified. This selected dichotomous rule correctly classified 28/33 R and 35/39 NR VSet pts (85% sensitivity, 90% specificity). ‘High risk’ pts had a significantly higher probability of R by 36 mo than ‘low risk’ pts: 88% PPV, 88% NPV; relative risk (RR) 7.0 (95% CI: 3.1, 16.1; p<0.0001). RR by stage (I/II) = 7.00/7.28 and tumor site (colon/rectum) = 8.75/4.50. Kaplan-Meier recurrence-free survival: hazard ratio 11.8 (95% CI: 4.5, 31.1; p<0.0001). NCCN Guidelines (V.2.2008) correctly classified 24/33 R and 15/39 NR VSet pts: 73% sensitivity, 38% specificity, 50% PPV, 63% NPV; RR 1.33 (95% CI: 0.74, 2.40; p=0.32). Conclusions: A GP derived 4-gene prognostic test using FFPE tumor tissue can differentiate early stage CRC pts at high versus low risk for R within 3y better than current NCCN Guidelines. [Table: see text] [Table: see text]