Multimodal CT radiomics predicts PD-1 inhibitor efficacy in advanced gastric cancer: a two-center validation study

TL;DR

A new model combining CT scans and clinical data can predict how well gastric cancer patients will respond to PD-1 inhibitors and chemotherapy.

Contribution

A novel clinical-radiomics model using CT radiomics and clinical features improves prediction of PD-1 inhibitor efficacy in gastric cancer.

Findings

The clinical-radiomics model using logistic regression achieved an AUC of 0.94 in internal and 0.85 in external validation.

CT radiomics features significantly improved prediction performance compared to clinical data alone.

The model reliably predicts response to PD-1 inhibitors combined with chemotherapy in advanced gastric cancer.

Abstract

In this study, we developed a multi-modal CT-based machine learning model to predict the response of gastric cancer (GC) patients to first-line chemotherapy combined with PD-1 inhibitors and performed external validation and multi-model comparisons.

We retrospectively analyzed the clinical data of 348 patients with GC who underwent immunotherapy. The patients were categorized into an internal validation cohort (center A, n = 272) and an external validation cohort (center B, n = 76). Pre-treatment clinical and CT radiomics features were extracted to develop three models: a clinical model, a radiomics model and a clinical-radiomics model. The classifiers included logistic regression (LR), linear support vector classification (Linear SVC), support vector machine, and random forest.

A total of 19 radiomics signatures and 5 clinical feature signatures were selected. In the radiomics model, the Linear SVC algorithm achieved an area under the receiver operating characteristic curve (AUC) of 0.88 and 0.76 in internal and external validation sets, respectively. In both the clinical model and the clinical-radiomics model, the LR algorithm demonstrated high and stable predictive performance in the internal (AUC = 0.89 and 0.94) and external validation datasets (AUC = 0.76 and 0.85). Among all models in the external validation dataset, the clinical-radiomics model utilizing LR outperformed all other classifiers.

The clinical-radiomics model, in combination with the LR algorithm, provides a reliable and effective method for predicting the early response of advanced GC patients treated with programmed cell death-1 (PD-1) inhibitors combined with chemotherapy.

CT radiomics and laboratory parameters were used to evaluate early prediction of response to PD-1 inhibitors combined with chemotherapy in patients with advanced gastric cancer. This clinical-radiomics model provides a novel approach to predict immunotherapy efficacy and prognosis.

Evaluating the efficacy of PD-1 inhibitors combined with chemotherapy in advanced gastric cancer using only clinical data is limited.Only some patients with advanced gastric cancer treated with the PD-1 inhibitors combined with chemotherapy achieved complete regression.This clinical-radiomics model showed good performance for predicting gastric cancer response to chemotherapy combined with PD-1 inhibitors.

Evaluating the efficacy of PD-1 inhibitors combined with chemotherapy in advanced gastric cancer using only clinical data is limited.

Only some patients with advanced gastric cancer treated with the PD-1 inhibitors combined with chemotherapy achieved complete regression.

This clinical-radiomics model showed good performance for predicting gastric cancer response to chemotherapy combined with PD-1 inhibitors.

Figures