To develop and validate a CpG-based classifier for preoperative discrimination of early and advanced-late stage colorectal cancer (CRC).

We identified an epigenetic signature based on methylation status of multiple CpG sites (CpGs) from 372 subjects in The Cancer Genome Atlas (TCGA) CRC cohort, and an external cohort (GSE48684) with 64 subjects by LASSO regression algorithm. A classifier derived from the methylation signature was used to establish a multivariable logistic regression model to predict the advanced-late stage of CRC. A nomogram was further developed by incorporating the classifier and some independent clinical risk factors, and its performance was evaluated by discrimination and calibration analysis. The prognostic value of the classifier was determined by survival analysis. Furthermore, the diagnostic performance of several CpGs in the methylation signature was evaluated.

The eight-CpG-based methylation signature discriminated early stage from advanced-late stage CRC, with a satisfactory AUC of more than 0.700 in both the training and validation sets. This methylation classifier was identified as an independent predictor for CRC staging. The nomogram showed favorable predictive power for preoperative staging, and the C-index reached 0.817 (95% CI: 0.753–0.881) and 0.817 (95% CI: 0.721–0.913) in another training set and validation set respectively, with good calibration. The patients stratified in the high-risk group by the methylation classifier had significantly worse survival outcome than those in the low-risk group. Combination diagnosis utilizing only four of the eight specific CpGs performed well, even in CRC patients with low CEA level or at early stage.

Our classifier is a valuable predictive indicator that can supplement established methods for more accurate preoperative staging and also provides prognostic information for CRC patients. Besides, the combination of multiple CpGs has a high value in the diagnosis of CRC.

开发和验证基于 CpG 的分类器，用于术前区分早期和晚期结直肠癌 (CRC)。

我们基于来自癌症基因组图谱 (TCGA) CRC 队列中的 372 名受试者的多个 CpG 位点 (CpG) 的甲基化状态以及通过 LASSO 回归算法具有 64 名受试者的外部队列 (GSE48684) 确定了表观遗传特征。使用源自甲基化特征的分类器建立多变量逻辑回归模型以预测CRC的晚期-晚期。通过结合分类器和一些独立的临床危险因素进一步开发了列线图，并通过判别和校准分析评估其性能。通过生存分析确定分类器的预后价值。此外，评估了甲基化特征中几个 CpG 的诊断性能。

基于 8 个 CpG 的甲基化特征区分早期与晚期晚期 CRC，在训练和验证集中均具有超过 0.700 的令人满意的 AUC。这种甲基化分类器被确定为 CRC 分期的独立预测因子。列线图对术前分期显示出良好的预测能力，在另一个训练集和验证集中，C 指数分别达到 0.817（95% CI：0.753-0.881）和 0.817（95% CI：0.721-0.913），校准良好。通过甲基化分类器分层的高风险组患者的生存结果明显低于低风险组的患者。即使在 CEA 水平低或处于早期阶段的 CRC 患者中，仅使用八种特定 CpG 中的四种的组合诊断也表现良好。

我们的分类器是一个有价值的预测指标，可以补充已建立的方法以实现更准确的术前分期，还可以为 CRC 患者提供预后信息。此外，多个CpGs的组合在CRC的诊断中具有很高的价值。