Keywords: overall survival, treatment effect, next line anti-cancer therapies
More and more effective cancer therapies are approved and become available to patients. In this era, an oncology clinical trial is faced with a familiar challenge at an elevated level. Upon progression, many patients may initiate other lines of anti-cancer therapies (NTX), e.g. chemotherapy, targeted therapy, and immunotherapy. The proportion of patients taking NTX, types of NTX and how many lines of NTX will differ between control and treatment arms. While overall survival (OS) is a gold standard for clinical benefit, the treatment effect on OS is confounded by NTX. If a treatment effectively delays or even prevents disease progression, it can be expected that more patients on the control arm will initiate NTX. As a result, the treatment effect on OS will be underestimated using the intent-to-treat analysis. Recently, NTX is conjectured as a possible reason for the non-significant result from JAVELIN Lung 200 (news release, Feb 15, 2018).
Morden et al and Watkinds et al discussed various methods to assess the true treatment effect when treatment switching is present. Naïve methods, like excluding switchers, censoring at switch, or using treatment as a time-varying covariate, are simple to implement but can suffer from potentially large estimation bias. Complex modeling approaches also exist, e.g. inverse-probability-of-censoring weighting (IPCW) and rank preserving structural failure time (RPSFT) model. Both models have untestable assumptions. Confounding introduced by NTX is more complex than that by treatment switching. Nevertheless, these articles provide a great starting point.
This round table discussion aims to provide an opportunity for statisticians in industry and regulatory agencies alike to share their experience and exchange thoughts on this challenging statistical problem.