Document resource
Background Reliable prognostic biomarkers are essential for guiding clinical decisions in patients treated with immune checkpoint inhibitors (ICIs). Current methods, such as RECIST and volumetric assessments, often lack sensitivity and clinical utility, contributing to uncertainty in clinical decisions. Deep learning analyses using comprehensive CT images may capture more complete prognostic imaging information beyond tumor size-based assessments. Expanding validation of such imaging biomarkers across diverse tumor types could enhance clinical utility and inform personalized management in ICI-treated patients.Methods Serial CT response score (Serial CTRS) is a fully automated deep learning biomarker that predicts OS by utilizing baseline and early-treatment CT scans. Serial CTRS was previously validated in advanced non-small cell lung cancer (aNSCLC), demonstrating superior OS predictions compared to conventional tumor size-based metrics. In this present study, we retrospectively applied Serial CTRS to 243 patients with advanced-stage cancers treated with PD-L1 ICIs at Providence Health System and West Cancer Center, including 87 metastatic renal cell carcinoma and other kidney cancer (RCC cohort), 125 metastatic melanoma and other skin cancers (melanoma cohort), and 31 extensive-stage small cell lung cancer (ES-SCLC cohort) cases. Patients who did not have available baseline and early treatment (28-120 days after ICI initiation) CT scans were excluded. Model thresholds predetermined in aNSCLC were uniformly applied to categorize patients into low, intermediate, and high Serial CTRS groups. Prognostic accuracy was assessed using Cox proportional hazards models, concordance index (C-index), and area under the receiver operating characteristic curve (AUROC) of landmark OS at 6, 12, and 24 months.Results Serial CTRS achieved a C-index of 0.77 (95% CI: 0.73–0.82) in the multi-indication cohort. Cohort-specific results were consistent: RCC cohort C-index 0.76 (95% CI: 0.67–0.84), melanoma cohort 0.76 (95% CI: 0.66–0.85), and ES-SCLC cohort 0.71 (95% CI: 0.61–0.81). 12-month OS AUROC ranged from 0.78–0.82 ( table 1). Kaplan-Meier analysis demonstrated clear OS stratification, with Serial CTRS high patients exhibiting significantly longer OS compared to low (HR: 10.54, 95% CI: 6.05–18.36). Intermediate versus high (HR: 4.49; 2.68–7.52) and low versus intermediate groups (HR: 2.60; 1.60–4.22) also demonstrated clinically relevant OS distinctions (figure 1).Conclusions Serial CTRS generalized effectively from aNSCLC to metastatic RCC, metastatic melanoma, and ES-SCLC, consistently stratifying OS without manual lesion annotations. This fully automated biomarker can facilitate objective prognostication, informing clinical decisions in oncology practices and clinical trials. Prospective studies and clinical workflow integration are planned to confirm its utility and impact.Ethics Approval The data collection was approved by the ethics committee/institutional review board (IRB) at each institution. Records were deidentified at the institution level per Health Insurance Portability and Accountability Act guidelines (US) or General Data Protection Regulation requirements (European Union). For data in the European Union, the patients were also notified that their deidentified data would be part of a study and were given the required time and opportunity to respond if they had any objections. On transfer, the data were quarantined and then reinspected by authorized personnel before ingestion to ensure compliance and that no PHI was present in the records.Abstract 1118 Table 1OS concordance index and 6, 12, 24-month AUROC for continuous serial CTRSAbstract 1118 Figure 1Kaplan-Meier OS plots stratified by 12-week Serial CTRS for all indications (A), metastatic RCC and kidney cancers (B), metastatic melanoma and skin cancers (C), and extensive-stage SCLC (D). Predetermined thresholds were uniformly applied to all data