Cost-Effectiveness Modeling Based on a Clinical Trial Re-analysis Using Multimodal Machine Artificial Intelligence Methodology: Predicting Which Patients May Benefit Most From the Addition of Tremelimumab to Durvalumab Plus Chemotherapy in First-Line...

Author(s)

Tristan Gonzalez, MSc1, Amanda Remorino, PhD2, Miguel Miranda, MS3, Vivian Peirce, PhD3, David Heidenwag, MS4, Colin Thierry, PhD5, Rucha Vadia, PhD6.
1AstraZeneca AG, Baar, Switzerland, 2AstraZeneca, Barcelona, Spain, 3AstraZeneca, Cambridge, United Kingdom, 4AstraZeneca, London, United Kingdom, 5Sophia Genetics, Bordeaux, France, 6Health Economics and Payer Evidence, AstraZeneca BeLux, Brussels, Belgium.
OBJECTIVES: The TRIDENT study was initiated to develop a multimodal artificial intelligence (AI) driven predictive model, integrating clinical, genomic and imaging data, to identify NSCLC patients who may derive greater OS benefit from the addition of tremelimumab (T) to durvalumab (D) and chemotherapy (Ctx) as 1st line metastatic treatment in the global, phase 3, randomised POSEIDON clinical trial. The model yielded genetic signatures identifying patients who may derive higher OS benefit (patients with conditional-average treatment effect scores above the median - designated “high-benefiters”). Amongst non-squamous NSCLC (NSCC) patients, high-benefiters experienced improved overall survival (OS) with T+D+Ctx vs. D + Ctx (hazard ratio [HR] 0.56; 95% CI: 0.33-0.97; n=172, compared to the TRIDENT NSCC population (HR 0.88; 95% CI: 0.60-1.11; n=345). Genomic features associated with higher benefit included EGFR, FGFR3, and CDKN2A wild-type status, and KRAS and STK11 mutations.Objectives(1) To estimate life-years (LYs) gained in high-benefiter patients receiving T+D+Ctx compared with D+Ctx; (2) To evaluate the feasibility and challenges of applying cost-effectiveness modeling to AI-derived evidence.
METHODS: A 3-state partitioned survival cost-effectiveness model was developed using time-to-event data from TRIDENT-defined high-benefiter NSCC patients treated with T+D+Ctx or D+Ctx. Time-to-event outcomes (OS, PFS and TTD) were extrapolated using survival models over a lifetime horizon, with models selected based on best statistical fit.
RESULTS: High-benefiter patients treated with T+D+Ctx were predicted to gain an additional 2.59 LYs compared to those treated with D+Ctx (4.4 LYs vs. 1.81 LYs). The results should be considered in view of both the strengths (meaningful improvement in survival) and inherent limitations of TRIDENT (restricted sample size and post-hoc stratification).
CONCLUSIONS: Multimodal AI-based stratification can identify patients most likely to benefit from specific treatment combinations, supporting more personalized strategies in oncology. Integrating this approach with cost-effectiveness analysis may enhance resource allocation, with positive implications for patient outcomes and healthcare system efficiency.

Conference/Value in Health Info

2025-11, ISPOR Europe 2025, Glasgow, Scotland

Value in Health, Volume 28, Issue S2

Code

MSR63

Topic

Economic Evaluation, Methodological & Statistical Research, Study Approaches

Topic Subcategory

Artificial Intelligence, Machine Learning, Predictive Analytics

Disease

Oncology

Your browser is out-of-date

ISPOR recommends that you update your browser for more security, speed and the best experience on ispor.org. Update my browser now

×