Overall survival (OS) remains the gold standard to assess treatment benefit in randomized controlled clinical trials (RCT) of advanced cancer. In first line metastatic breast cancer (1 L MBC) where survival typically is several years instead of months, it is increasingly seen that OS-treatment effects dilute over time, with later line therapies strongly influencing OS analyses. We investigate truncated OS (i.e., truncating deaths and follow-up beyond a certain time point from randomization) as a potential alternative endpoint to OS.
We reanalyze data from two RCTs in 1 L MBC to illustrate how OS results can be driven by early treatment effects. We flexibly model the hazard ratio (HR) according to time since randomization. We conduct simulations to characterize the impact of number of events, sample-size, and recruitment on the power of the log-rank test for OS and for truncated OS considering truncation at 12, 18, or 24 months in presence of non-proportional hazards (non-PH).
The impact of non-PH on the power of a log-rank test for OS may be severe and the power for a truncated OS analysis is higher in situations of non-PH. Statistically significant OS benefit is unlikely to be demonstrated with a log-rank test unless treatment effect is very strong, long-lasting, or only short-term follow-up is included.
Compared with the classical OS endpoint, truncated OS may provide the treating physicians, prescribers, and payers with a more accurate and less biased estimate of treatment benefit more directly attributable to the treatment under investigation.