>> Anyway, the question is why do we have this disconnect between models applied to both arms, and the real time timing of the first look.
Using mOS alone of each arm to project events for a trial just finished enrollment is a non-starter because the underlying assumption is that KM curves are separating at the beginning when they usually don't separate much at least for a few months in trials like this. Earlier phase trials that showed OS curves separate from the beginning are usually not replicable in larger ph3 trials. For example, SNTA almost 4 months mOS advantage from ph2 trial disappeared completely in ph3 trial resulting futility at 1st interim. Event rate on both arms shouldn't differ that much in early interim, see Opdivo trial, event rate didn't differ until after 6 months. Thus models used here to produce 9:1 events at 1st interim have practically zero value at this point. I would put reasons for the disconnect in the following order:
. Enrolment way "backloaded"
. Dropouts (or just slow event data collection) in the E-EU sites
. Placebo arm doing better than guessed
In previous 2nd line NSCLC trials, few effective subsequent treatments after progression were available. Very few patients in Opdivo trial went on to immunotherapy after progression even though 50% on both arms went on to have subsequent treatments. However for ongoing NSCLC trials, it is different as both Opdivo and Keytruda are approved for NSCLC for a few months now, thus it is very likely to have effect on both event rate and dropout, depends on how many patients enrolled are from US. How much is the impact, don't know until full results. Even though official dropout in Opdivo trial was very low, 4 on each arm, but 22 on docetaxel arm withdrew consent after randomization thus didn't receive docetaxel treatment, and very high percentage (20%) discontinued treatment due to AEs on docetaxel arm as well. Only extremely well run trial could keep almost all of these patients in follow up for OS.