Abstract 2037: Early identification of long-term survivors on tremelimumab plus durvalumab and chemotherapy in POSEIDON from pre-treatment CT images

Abstract Background: The combination of CTLA-4 and PD-(L)1 inhibitors either with or without chemotherapy is an approved treatment for metastatic non-small-cell lung cancer (mNSCLC). Identifying which patients are most likely to derive additional benefit from these regimens, as compared anti-PD-1 alone or in combination with chemotherapy (CTx) would help optimize clinical management of these patients. Methods: Pre-treatment baseline Computed Tomography (CT) images of the thorax from patients treated with durvalumab plus tremelimulab (D+T; n=647) as part of the MYSTIC and NEPTUNE studies were split into a training (n=557) and a validation cohort (n=90). A 3D Deep Learning model with a DenseNet architecture was trained to predict patients with overall survival (OS) longer or shorter than the median (∼1 year), and then optimized using the validation set. The final model (named RaDx) was then applied de-novo to the non-squamous mNSCLC cohort from POSEIDON. Results: All reported hazard ratios (HR) are comparing RaDx+ and RaDx- groups unless specified otherwise. In the validation cohort, the median OS (mOS) in the RaDx+ (n=38) and the RaDx- (n=52) groups was 21 and 6.3 months respectively with HR=0.44 (0.27-0.73). In the anti-CTLA-4 containing arm of POSEIDON (D+T+CTx; n=187), the mOS for RaDx+ (n=77) and RaDx- (n=110) groups was 26.6 and 10.8 months respectively with HR = 0.44 (0.3-0.64). In the durvalumab and chemotherapy arm (D+CTx; n=182), the mOS for the RaDx+ (n=71) and RaDx- (n=111) groups was 17.2 and 11.2 months respectively with HR = 0.68 (0.48-0.97). In the CTx only arm (n=178), the mOS for the RaDx+ (n=73) and RaDx- (n=105) groups was 15.1 and 11.6 months respectively with HR = 0.82 (0.59-1.13). Within the RaDx+ sub-group a comparison of D+T+CTx to D+CTx patients favoured D+T+CTx with HR = 0.63 (0.42-0.96). Conclusions: We demonstrated feasibility that a Deep Learning model using baseline CT images may be used to identify a sub-group of patients for whom the addition of tremelimumab to chemo-immunotherapy may bring additional benefit. Such a biomarker could have the potential to guide optimal treatment selection in this 1L mNSCLC population. Citation Format: Kedar Patwardhan, Harish RaviPrakash, Qin Li, Ross Stewart. Early identification of long-term survivors on tremelimumab plus durvalumab and chemotherapy in POSEIDON from pre-treatment CT images [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 2037.