Development and validation of serological dynamic risk score to predict outcome in gastric cancer with adjuvant chemotherapy: a multicentre, longitudinal, cohort study.

ABSTRACT

Background: Baseline serological biomarkers have the potential to predict the benefits of adjuvant chemotherapy in patients with gastric cancer. However, the fluctuating nature of postoperative recurrence risk makes precise treatment challenging. We aimed to develop a risk score in real-time predicting outcomes for postoperative GC patients using blood chemistry tests. Materials and methods: This was a retrospective, multicentre, longitudinal cohort study from three cancer centres in China, with a total of 2737 GC patients in the pTNM stage Ib to III. Among them, 1651 patients with at least two serological records were assigned to the training cohort. Model validation was carried out using separate testing data with area under curve (AUC). The least absolute shrinkage and selection operator (LASSO) and random forest-recursive feature elimination (RF-RFE) algorithm were used to select the parameters. Results: The Cox regression model derived six risk factors to construct a composite score (low-risk 0-2 score; high risk 3-6 score), including CEA, CA125, CA199, haemoglobin, albumin, and neutrophil to lymphocyte ratio. The risk score accurately predicted mortality in 1000-time bootstrap (AUROCs0.658; 95% CI 0.645, 0.670), with the highest AUROC (0.767; 95% CI 0.743, 0.791) after 1 year since the gastrectomy. In validation dataset, the risk score had an AUROC of 0.586 (95% CI 0.544, 0.628). Furthermore, patients with high risk at 1 month derived significant clinical benefits from adjuvant chemotherapy (P for interaction <0.0001). Compared with the low-low-low risk group, the low-low-high risk group of the long-term state chain (risk state at baseline, 6 months, 1 year) had the worse OS (HR, 6.91; 95%CI 4.27, 11.19) and DFS (HR, 7.27; 95%CI 4.55, 11.63). Conclusion: The dynamic risk score is an accurate and user-friendly serological risk assessment tool for predicting outcomes and assisting clinical decisions after gastrectomy.