肿瘤诊断模型+风险预后模型双模型

Tumour-infiltrating immune cells are a source of important prognostic information for patients with resectable colon cancer. We developed a novel immune model based on systematic assessments of the immune landscape inferred from bulk tumor transcriptomes of stage I–III colon cancer patients. The “Cell type Identification By Estimating Relative Subsets Of RNA Transcripts (CIBERSORT)” algorithm was used to estimate the fraction of 22 immune cell types from six microarray public datasets. The random forest method and least absolute shrinkage and selection operator model were then used to establish immunoscores for diagnosis and prognosis. By comparing immune cell compositions in samples of 870 colon cancer patients and 70 normal controls, we constructed a diagnostic model, designated the diagnostic immune risk score (dIRS), that showed high specificity and sensitivity in both the training [area under the curve (AUC) = 0.98, p < 0.001] and validation (AUC 0.96, p < 0.001) sets. We also established a prognostic immune risk score (pIRS) that was found to be an independent prognostic factor for relapse-free survival in every series (training: HR 2.23; validation: HR 1.65; entire: HR 2.01; p < 0.001 for all), which showed better prognostic value than TNM stage. In addition, integration of the pIRS with clinical characteristics in a composite nomogram showed improved accuracy of relapse risk prediction, providing a higher net benefit than TNM stage, with well-fitted calibration curves. The proposed dIRS and pIRS models represent promising novel signatures for the diagnosis and prognosis prediction of colon cancer.