Authors

Jeng-Kai Jiang, Hung-Shin Lin, Jen-Kou Lin, Yu-Chung Wu, Teh-Ying Chou, Nien Wei; Taipei Veterans General Hospital and National Yang-Ming University, Taipei, Taiwan; Taipei Veterans General Hospital, Taipei, Taiwan; Auspex Diagnostics, Taipei, Taiwan

Background

Combination chemotherapy using fluorouracil, leucovorin with or without oxaliplatin (5-FU and FOLFOX) are standard treatment for locally advanced colon cancer. However, about half of the patients developed recurrence. The objective of this study was to use gene profiling to predict treatment efficacy for colon cancer patients (stages 1-3). We developed three prediction models: the first determines the likelihood of recurrence after curative resection while the other two models determine the efficacy of 5-FU and FOLFOX.

Methods

Formalin-fixed paraffin-embedded tissues from 513 patients treated during 2003 to 2011 at Taipei Veterans General Hospital (80% of them are treated between 2005 to 2009) were analyzed for gene expression using Nugen Ovation kits and Affymetirix’s ST arrays. The gene data was analyzed using supervised clustering K-nearest-neighbor method. The performance of the models was evaluated using a Hazard Ratio (HR) for disease-free survival.

Results

In model one, for prediction of recurrence, the HR of non-recurrence vs. recurrence is 2.758 (P<0.0001). In model two, for determination of 5FU efficacy, the HR of responder vs. non-responder is 3.212 (P=0.0001). In model three, for determination of FOLFOX efficacy, the HR of responder vs. non-responder is 3.604 (P< 0.0001). Although these three models were developed separately, the cross predictions among models are consistent. For example: the non-responders predicted by the 5-FU or FOLFX models are also predicted to be recurrent patients by the recurrence prediction model. Patients predicted to be FOLFOX non-responders are also identified as 5-FU non-responders by the 5-FU treatment model. While most patients were predicted to respond similarly to 5-FU or FOLFOX treatment, our models were also able to identify a subset of patients that would fail 5-FU treatment but benefit from FOLFOX therapy.

Conclusions

A consistent three-model-prediction has been developed that provides valuable indictors for colon cancer treatment.