Document resource
Background ICIs have emerged as an effective treatment option for NSCLC, offering durable survival benefits even in the first-line setting. However, variability in individual response to immunotherapy remains a major challenge. Reliable predictive biomarkers remain an unmet need. Radiomics, the comprehensive quantification of tumor phenotypes through automated extraction of quantitative image features, has demonstrated prognostic value in cancer imaging. 1 Recent studies have shown that radiomic biomarkers can predict immunotherapy response noninvasively in patients with advanced cancer.2 We aimed to develop a non-invasive, radiomic signature to predict overall-survival and identify patients at high risk of mortality.Methods Baseline CT scans from 95 consecutively treated stage IV NSCLC patients were retrospectively analysed. Following semi-automatic primary-tumour segmentation, 533 deep-radiomic features using handcrafted segmentation that capture first-order intensity, three-dimensional morphology and diverse texture matrices were extracted, concatenating with clinical covariates (age, sex, ECOG status, NTL ratio, number of metastatic organs and PD-L1 expression). Feature selection of ANOVA filtering and model construction were embedded in a nested five-fold cross-validation for ensemble learners. Model performance was assessed for 12- and 36-month OS and PFS using the mean cross-validated area under the AUROC with bootstrap confidence intervals. For interpretability, patients were stratified into high- and low-risk groups according to optimised cut-points of the cross-validated risk score, and survival distributions were compared with Kaplan-Meier analysis.Results Among stage IV 95 patients, 33 patients(32.0%) were diagnosed with brain metastases prior to initiating immunotherapy. During a median follow-up of 14 months (IQR 5-30), 80 deaths were recorded in the study (median age 66 years, IQR 59.5-73.5). The final radiomic ensemble model demonstrated an AUC of 0.79 (95% CI: 0.70-0.88) for 12-month OS ( figure 1) and 0.66 (0.53–0.79) for 36-month OS (figure 2). Corresponding PFS performance was 0.68 (0.56–0.81) at 12 months and 0.72 (0.58–0.84) at 36 months (figure 2). Kaplan-Meier curves demonstrated clear risk discrimination: for the high-risk cohort had a median survival of 9 mo, whereas the low-risk cohort had a median survival of 21 mo (HR 2.08, 95% CI 1.32–3.29; p < 0.001) (figure 1). Similar separation was observed for 36-month PFS (HR 2.12, p < 0.001).Conclusions A CT-based radiomic signature restricted to stage IV NSCLC provides robust, time-specific prognostic information and accurately separates patients with markedly different survival trajectories under ICI therapy. Because the model relies solely on standard imaging and machine-learning analysis, it can be seamlessly integrated into clinical workflows to guide personalized immunotherapy and enrich future clinical trials, pending prospective validation.References Aerts HJWL, Velazquez ER, Leijenaar RTH, Parmar C, Grossmann P, Carvalho S, Bussink J, Monshouwer R, Haibe-Kains B, Rietveld D, Hoebers F, Rietbergen MM, Leemans CR, Dekker A, Quackenbush J, Gillies RJ, Lambin P. Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. Nat Commun. 2014;5:4006.Trebeschi S, Drago SG, Birkbak NJ, Kurilova I, Calin AM, Delli Pizzi A, Lalezari F, Lambregts DMJ, Rohaan MW, Haanen JBAG, Blank CU, Beets-Tan RGH. Predicting response to cancer immunotherapy using noninvasive radiomic biomarkers. Ann Oncol. 2019; 30:998–1004.Ethics Approval This study was approved by the Institutional Review Board Committee Northwestern University, Chicago, IL, USA (No. STU0020711; Approval date: 04/06/2018). Consent was waived for its retrospective manner and the was no identifiable risk for patients included in this analysis.Abstract 1120 Figure 112-Month OS ROC and Kaplan-Meier curves. The left plot shows the ROC curve for 12-month OS prediction with an AUC of 0.792 (95% CI: 0.701-0.880). The right plot illustrates Kaplan-Meier curves, distinguishing low and high-risk patients (p < 0.001)Abstract 1120 Figure 2Time-specific survival prediction performance. The bar chart displays the AUC for 12-month OS, 36-month OS, 12-month PFS, and 36-month PFS, with 12-month OS showing the best performance (AUC = 0.792)