OS and PFS are commonly correlated with response upon RECIST criteria. Whether a more dynamic approach considering overall and individual lesions trajectories may be of predictive/prognostic value is worth investigating.
Target lesions size (TL) from AXIS trial (NCT00678392: axitinib (A) vs sorafenib (S)) at baseline (bslTL) and subsequent time-points up to 24 months were retrospectively analyzed. TL were grouped by major organs. Relationship between OS/PFS and TL was assessed using mean-trajectories and joint-models (JM) with Cox/Weibull regression (C/W-R) including time to treatment interaction, with/without non-linear term. Mean trajectories obtained from unsupervised longitudinal clustering (ULC) were subject to C/W-R including S/A effect, age and IMDC risk group.
Among 713 eligible pts (A:359, S:354), overall TL trajectories of A and S did not differ for the 1st 12 months, whereas individual TL trajectories strongly varied with TL localization: largest difference between A and S were observed for bone and liver, no difference for lungs. Upon ULC, 3 clusters exhibiting different patterns were characterized, cl1: bslTL ≈ 50mm not time-varying, cl2: bslTL ≈ 150mm + decreasing rate = 5mm/month and cl3: bslTL ≈ 250mm + decreasing rate = 15mm/month. Surprisingly, cluster 3 exhibited the better PFS: HR = 0.51 (0.31-0.82) and OS: HR = 0.54 (0.29-1.02). Upon JM, PFS and OS were associated with TL trajectories, respectively p = 0.0004 and p
Unlike overall TL, individual TL different trajectories were exhibited by A and S. Our model identified 3 clusters with different predictive and prognostic characteristics.
Clinical trial identification
Legal entity responsible for the study
All authors have declared no conflicts of interest.