Development and validation of metabolic scoring to individually predict prognosis and monitor recurrence early in gastric cancer: A large-sample analysis.

PURPOSE

To develop and validate a simple metabolic score (Metabolic score, MS) for use in evaluating the prognosis of gastric cancer (GC) patients and dynamically monitor for early recurrence.

METHODS

We retrospectively collected general clinicopathological data of patients who underwent radical gastrectomy for GC between September 2012 and December 2017 in the Department of Gastric Surgery of the Fujian Medical University Union Hospital. Using a random forest algorithm to screen preoperative blood indicators into the Least absolute shrinkage and selection operator (LASSO) model, we developed a novel MS to predict prognosis.

RESULTS

Data of 1974 patients were used to develop and validate the model. Total cholesterol (TCHO), bilirubin (TBIL), direct bilirubin (DBIL), and 15 other metabolic indicators had significant predictive value for the prognosis using the random forest algorithm. In the overall population, 533 patients (27.0%) had high and 1441 (73%) had low MS status. High MS status was related to tumor progression. The KM curves of 3-year OS and RFS for training set patients showed low MS had a better prognosis than high MS (OS: 79.4% vs 59.7%, P < 0.001; RFS: 76.0% vs 56.2%, P < 0.001).

CONCLUSIONS

We have developed and validated MS to predict the long-term survival of GC patients and allow early monitoring of recurrence. This will provide physicians with simple, economical, and dynamic tumor monitoring information.