Prediction of Distant Metastasis Using CT/PET-Based q Radiomics in Non-Small Cell Lung Cancer: A Systematic Review and Meta-Analysis

Document Type

Conference Proceeding

Publication Date

10-23-2025

Publication Title

J Thorac Oncol

Keywords

case report, conference abstract, distant metastasis, human, male, meta analysis, non small cell lung cancer, observational study, positron emission tomography-computed tomography, prediction, predictive model, Preferred Reporting Items for Systematic Reviews and Meta-Analyses, radiomics, randomized controlled trial, retrospective study, systematic review

Abstract

Introduction: Non-small cell lung cancer (NSCLC) is associated with high rates of distant metastasis (DM), which significantly impacts patients' outcomes. The use of CT/PET-based radiomics has emerged as a promising tool to predict DM, potentially offering more accurate predictions compared to traditional clinical models. We aim to eval uate and compare the predictive performance of radiomics and clinical models in predicting DM in NSCLC patients. Methods: We conducted a systematic review and meta-analysis following the PRISMA guidelines through PubMed and Scopus databases for studies on radiomics in NSCLC. The protocol was registered on PROSPERO (CRD42025628953). Observational studies and randomized clinical trials on NSCLC patients who underwent CT/PET-based radiomics with distant organ metastasis were included. Non-English articles, reviews, case reports, and animal studies were excluded. Primary outcome included accuracy of CT/PET-based radiomics in predicting DM in NSCLC. Pooled analysis was performed using the inverse vari ance weighting meta-analysis of single means on evaluation metrics including the area under curve (AUC), and the concordance index (Cindex). Secondary outcomes included the comparison of radiomics vs. clinical models in DM prediction. Results: Seventeen studies, including 1,550 NSCLC patients, were analyzed. Among them, 559 patients had distant metastasis. The mean age was 69.6 ± 7.7 years, and 733 were males. Stage I, II, and III diseases were reported in 450, 316, and 291 patients, respectively. Six studies assessed DM predic tion using the C-index, and 11 used the AUC. For radiomics models evaluated with the AUC, low heterogeneity was observed (I2 = 42%, p = 0.07), with a pooled AUC of 0.73 (95% CI: 0.70-0.77) using a fixedeffect model. Radiomics models evaluated with the C-index showed no heterogeneity (I2 = 0%, p = 0.83), with a pooled C-index of 0.65 (95% CI: 0.61-0.70). Clinical models assessed using the AUC displayed high heterogeneity (I2 = 90%, p < 0.01), with a pooled AUC of 0.70 (95% CI: 0.61-0.79) using a random-effect model. In contrast, clinical models evaluated with the C-index showed no heterogeneity (I2 = 0%, p = 0.61), with a pooled C-index of 0.60 (95% CI: 0.55-0.65). Conclusions: Our results indicate that radiomics models outperformed clinical models in predicting DM in NSCLC patients, as evidenced by higher pooled AUC and C-index values. Radiomics models also exhibited low heterogeneity across studies, suggesting consistent and reproducible performance. In comparison, clinical models showed greater vari ability and slightly lower predictive accuracy. These findings suggest that radiomics has the potential to serve as a reliable tool for DM prediction in NSCLC, offering more consistent performance than clinical models.

Volume

20

Issue

10

First Page

S279

Find It @ Sladen

Share

COinS