Longitudinal assessment of ROPRO as an early indicator of overall survival in oncology clinical trials: a retrospective analysis

BackgroundThe gold standard to evaluate treatment efficacy in oncology clinical trials is Overall Survival (OS). Its utility, however, is limited by the need for long trial duration and large sample sizes. Thus methods such as Progression-Free Survival (PFS) are applied to obtain early OS estimates across clinical trial phases, particularly to decide on further development of new molecular entities. Especially for cancer-immunotherapy, these established methods may be less suitable. Therefore, alternative approaches to obtain early OS estimates are required. In this work, we present a first evaluation of a new method, {Delta}Risk. {Delta}Risk uses the ROPRO, a state-of-the-art pan-cancer OS prognostic score, or DeepROPRO to predict OS benefit by measuring the patients improvement since baseline. Patients and methodsWe modeled the {Delta}Risk using Joint Models and tested whether a significant {Delta}Risk decrease correlated with OS improvement. We studied this hypothesis by comparing classical OS analysis against {Delta}Risk in a retrospective analysis of 12 real-world data emulated clinical trials, and 3 additional recent phase III immunotherapy clinical trials. ResultsOur new {Delta}Risk method correlated with the final OS readout in 14 out of 15 clinical trials. The {Delta}Risk, however, identified the treatment benefit up to seven months earlier than the OS log-rank test. Additionally, in two immunotherapy trials where PFS would have failed as an early OS estimate, the {Delta}Risk correctly predicted the treatment benefit. ConclusionsWe introduced a new method, {Delta}Risk, and demonstrated its correlation with OS. In retrospective analysis, {Delta}Risk is able to identify OS benefit earlier than standard methodology, and we show examples of lung cancer trials, where it maintains its predictive relevance whereas PFS does not correlate with OS. {Delta}Risk may prove useful for early decision support resulting in reduced need of resources. We also show the potential of {Delta}Risk as a candidate to define surrogate endpoints. To this purpose, more methodological work and further investigation of treatment-specific performance will be done in the future.