Abstract
BACKGROUND: Cancer recurrence is an important predictor of survival outcomes in patients with colorectal cancer-associated liver metastasis (CRLM), who undergo radical hepatectomy. Therefore, identification of patients with the greatest risk of recurrence is critical for developing a precision oncology strategy that might include frequent surveillance (in low-risk patients) or a more aggressive treatment approach (in high-risk patients). We performed genome-wide expression profiling, to identify and develop a transcriptomic signature for predicting recurrence in patients with CRLM. METHODS: We analysed a total of 383 patients with CRLM, including 63 patients from a publicly available data set (the NCBI's Gene Expression Omnibus with accession number GSE81423). and 320 patients from whom surgical specimens were collected for independent training (n = 169) and validation (n = 151) of identified biomarkers. Using Cox's proportional hazard regression analysis, we evaluated the clinical significance of the identified gene signature by comparing its performance with several key clinical factors. RESULTS: We identified a six-gene panel that robustly categorised patients with recurrence in the discovery (area under the curve (AUC) = 0.90). We showed that the panel was a significant predictor of recurrence in the clinical training (AUC = 0.83) and validation cohorts (AUC = 0.81). By combining our panel with key clinical factors, we established a risk-stratification model that emerged as an independent predictor of recurrence (AUC = 0.85; univariate: hazard ratio (HR) = 4.34, 95% confidence interval (CI) = 2.71-6.93, P < 0.001; multivariate: HR = 3.40, 95% CI = 1.76-6.56, P < 0.001). The stratification model revealed recurrence prediction in 89% of high-risk group and non-recurrence in 62% of low-risk group. CONCLUSIONS: We established a novel transcriptomic signature that robustly predicts recurrence, which has significant implications for the management of patients with CRLM.