AUTHOR=Fan Jin , Liu Yanlong , Cai Xin , Wang Jingwen , Guo Rui , Ji Yuan , Li Chao , Xu Ye , Li Xinxiang , Zhang Chundong , Zhang Rui , Zhu Ji , Cai Sanjun TITLE=A Novel Prognostic Model Incorporating Carcinoembryonic Antigen in 3-Week or Longer Postoperative Period for Stage III Colon Cancer: A Multicenter Retrospective Study JOURNAL=Frontiers in Oncology VOLUME=10 YEAR=2020 URL=https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2020.566784 DOI=10.3389/fonc.2020.566784 ISSN=2234-943X ABSTRACT=Background

The prognostic stratification of colon cancer using only the tumor-node-metastasis (TNM) stage has some limitations. We sought to increase the accuracy of stratifying patients with stage III colon cancer by constructing a prognostic model combining carcinoembryonic antigen (CEA) with TNM.

Methods

We retrospectively analyzed the data generated from stage III colon cancer patients who had early postoperative CEA measurement from 21 to 100 days after surgery from 2006 to 2017. CEA value was processed using restricted cubic splines (RCS) method. The prognostic model was developed using cox proportional hazards regression.

Results

The time later than 20 days after surgery was optimal for measuring CEA, which was determined by comparing the prognostic value for preoperative and postoperative CEA (N = 2,049), and by evaluating the relationship between the hazard ratio (HR) and postoperative CEA measuring time. Postoperative CEA, T stage and N stage were selected into the final model, and the mean integrated-AUC (iAUC) was 0.78 with 1,000 × bootstrap resampling, which was higher than the model using only T and N stages (TN model; mean iAUC, 0.66). The net reclassification improvement (NRI) was 15% when compared with TN model. Patients could be divided into high and low risk groups by the model, and 3-year disease-free survival (DFS) were 53.7% and 87.0%, respectively (HR, 4.30; 95% CI, 2.65 to 6.96; P < 0.001). Similar results were found in the validation set.

Conclusions

Stage III colon cancer could be stratified more accurately using the new prognostic model combining postoperative CEA with T and N stage.