Immunotherapy could become the standard treatment in many types of cancer in the near future. Treatment effects on time-to-event endpoints are described by proportional-hazards models (PH). PH assumption leads to powerful tests when correct but in the context of IT, it can be poor, leading to significant power losses. More, non-PH models work poorly in PH situations and require some knowledge of the form of non-PH which is not available. We propose a method that works well in both situations applicable to IT.
Data from nivolumab pivotal clinical trials (melanoma: NCT01721772, NSCLC: NCT01642004) were analyzed to investigate the impact of non-PH on OS. These studies relied on a hypothesis of PH situation but did not investigate the impact of its possible violation on final results. We used a general multivariate non-PH model in which very broad families of situations can be described. This allowed construction of a test based on integrated Brownian motion (O'Quigley 2003).
Melanoma trial exhibited PH whereas NSCLC did not. Our test was applied in both trials and exhibited an almost identical power to that of the log-rank test under PH situations but had typically much greater power under non-PH situations. For the NSCLC data, the test showed the data to be well modeled by a PH model (p
Immunotherapy survival curves exhibit peculiarities which may violate the underlying PH assumption. We analyzed nivolumab pivotal trials for OS and demonstrated usefulness of our test. This test may be suitable for any condition encountered in immunotherapy trials and is not dependent on any PH assumption.
Clinical trial identification
Legal entity responsible for the study
All authors have declared no conflicts of interest.