利用外周血临床实验室组学数据的动态变化预测晚期胃癌免疫检查点抑制剂疗效的机器学习模型的开发与验证：一项回顾性多中心队列研究

Development and validation of a machine learning model for predicting immune checkpoint inhibitor efficacy in advanced gastric cancer using dynamic changes in peripheral blood clinlabomics data: a retrospective multicenter cohort study.

作者信息

Nie Shulun, Song Shuyi, Xu Qian, Dai Xin, Liu Aina, Sun Meili, Cong Lei, Liang Jing, Liu Zimin, Lv Jing, Li Zhen, Zhang Jinling, Cao Fangli, Qu Linli, Liu Haiyan, Yue Lu, Zhai Yi, Li Song, Liu Lian

机构信息

Department of Medical Oncology, Qilu Hospital, Cheeloo College of Medicine, Shandong University, Jinan, Shandong, China.

Department of Medical Oncology, Shandong Provincial Hospital of Traditional Chinese Medicine, Jinan, Shandong, China.

出版信息

Gastric Cancer. 2025 Sep 7. doi: 10.1007/s10120-025-01655-1.

DOI:10.1007/s10120-025-01655-1

PMID:40914924

Abstract

BACKGROUND

Immune checkpoint inhibitors (ICIs) play a pivotal role in the treatment of advanced gastric cancer (GC). However, the biomarkers used to predict ICI efficacy are limited due to their reliance on single or static tumor characteristics. This study aims to develop a machine learning (ML) model that incorporates dynamic changes in clinlabomics data to optimize the predictive accuracy of ICI efficacy.

METHODS

This multicenter, retrospective study utilized nine ML to construct the model. Participants were further stratified into low-risk and high-risk groups based on the predicted efficacy of ICI. Kaplan-Meier survival curves and RNA-sequencing were used for differential analysis.

RESULTS

This study enrolled 377 patients with advanced GC who underwent first-line ICI treatment across eleven hospitals between January 2018 and May 2023. Among them, 220 patients from Qilu Hospital of Shandong University were selected for the development model. The remaining ten hospitals contributed to two external test cohorts. Ten dynamic clinlabomics features were identified. The XGBoost demonstrated optimal performance in predicting ICI response, achieving an AUC of 0.863 in the training cohort, and 0.790-0.842 in the validation and two external cohorts. Notably, the model exhibited strong predictive capabilities compared to single point-in-time and previously proposed model. In the subgroup analysis, the low-risk subtype demonstrated a significantly improved prognosis and exhibited characteristics of "hot tumors". A web tool was generated: https://ici-therapeutic-efficacy-predictor-ztwwfwek2uckbmhxlnsayq.streamlit.app/ .

CONCLUSIONS

The dynamic clinlabomics model can effectively predict the ICI efficacy in advanced GC. The model was validated using multicenter data and provides new evidence to optimize treatment decisions.

摘要

背景

免疫检查点抑制剂（ICI）在晚期胃癌（GC）治疗中发挥着关键作用。然而，用于预测ICI疗效的生物标志物因依赖单一或静态肿瘤特征而有限。本研究旨在开发一种机器学习（ML）模型，该模型纳入临床实验室组学数据的动态变化，以优化ICI疗效的预测准确性。

方法

这项多中心回顾性研究利用九种机器学习方法构建模型。根据ICI的预测疗效，将参与者进一步分为低风险和高风险组。采用Kaplan-Meier生存曲线和RNA测序进行差异分析。

结果

本研究纳入了2018年1月至2023年5月期间在11家医院接受一线ICI治疗的377例晚期GC患者。其中，山东大学齐鲁医院的220例患者被选入开发模型。其余十家医院贡献了两个外部测试队列。确定了十个动态临床实验室组学特征。XGBoost在预测ICI反应方面表现出最佳性能，在训练队列中的AUC为0.863，在验证队列和两个外部队列中的AUC为0.790 - 0.842。值得注意的是，与单点时间模型和先前提出的模型相比，该模型表现出强大的预测能力。在亚组分析中，低风险亚型显示出显著改善的预后，并表现出“热肿瘤”的特征。生成了一个网络工具：https://ici-therapeutic-efficacy-predictor-ztwwfwek2uckbmhxlnsayq.streamlit.app/ 。