The Lymph Node Ratio as a Predictive Biomarker for Individualized Benefit from Adjuvant Chemotherapy in Gastric Cancer: A Retrospective Cohort and Causal Machine Learning Study.

Authors

Wen-Qi Hong, Ren-Hao Hu, Xiao-Hua Jiang, Shun Zhang

Published in

Journal of gastrointestinal cancer. Volume 57. Issue 1. Aug 22, 2026. Epub Aug 22, 2026.

Abstract

Optimizing adjuvant chemotherapy (AC) for gastric cancer (GC) remains challenging due to patient heterogeneity. While the lymph node ratio (LNR) is a known prognostic factor, its role in predicting individualized AC benefit remains underexplored. This study aimed to leverage causal machine learning to explore LNR's role for personalized treatment.
We conducted a retrospective cohort study of 2,748 patients undergoing radical gastrectomy (2007-2017, re-staged by AJCC 8th edition). While the full cohort provided a demographic overview, analytic models focused on untreated Stage IB patients (n = 325) for prognostic factors, and Stage II-III patients (n = 825) for AC benefit estimation using a Causal Forest model with out-of-bag (OOB) predictions. Propensity score matching (PSM) was employed to mitigate treatment allocation bias.
AC benefit was highly heterogeneous. In Stage IB, lymphovascular invasion (LVI) and elderly age were independent prognostic factors. Strikingly, the Causal Forest model (Area Under the Uplift Curve = 31.08) identified LNR as the most dominant predictor of AC benefit. Subgroup Cox interaction analysis within the PSM cohort (n = 434) confirmed a highly significant threshold effect (P for interaction < 0.001): AC significantly reduced mortality in the High LNR (> 0.25) group (HR = 0.41, P = 0.016), while showing potential harm in the Low LNR (< 0.1) group (HR = 2.58, P = 0.015). The top 20% of predicted beneficiaries achieved an Absolute Risk Reduction (ARR) of 21.59%, corresponding to a Number Needed to Treat (NNT) of 4.63.
LNR is identified as a robust predictive biomarker for AC benefit in GC. This exploratory causal inference framework can help personalize treatment decisions, representing a valuable approach to complement clinical guidelines. To facilitate clinical application, an exploratory web-based decision support tool was developed.

PMID:
42631885
Bibliographic data and abstract were imported from PubMed on 23 Aug 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.