Two forums run under the FDA Oncology Center of Excellence’s Project SignifiCanT — one in June, one in September 2025 — dropped their summary reports close enough together to read as a pair. Both involve endpoints that feel settled until they aren’t.
When SOC shifts under your feet
The June forum, co-hosted with LUNGevity Foundation, addressed what happens when a new therapy is approved mid-trial and your control arm quietly becomes the inferior option. The estimand framing matters here: “add-on” designs, where new SOC is layered into both arms, present different analytical problems than designs where the new therapy enters only the control. The latter — the harder case — offers limited viable options for pooled or stratified analysis without strong, often untenable assumptions, including proportional hazards and absence of time trends from non-concurrent data.
The panel’s honest conclusion: in many scenarios, a full trial restart is the most defensible path. That is an uncomfortable message for sponsors who have spent years and tens of millions on a running study, but regulators from FDA, Health Canada, MHRA, TGA, and several others were all in the room and appear to be converging on the same view. Lung-MAP’s full redesign after nivolumab approval was cited as a real-world precedent. The practical implication now — before the guidance arrives — is pre-specifying contingency plans for SOC evolution in your protocol and SAP. If you haven’t done that for your ongoing oncology RCTs, the forum’s output is a useful prompt.
Duration of Response: a familiar number with a methodological debt
The September forum took on Duration of Response, an endpoint that appears on nearly every oncology label and is rarely examined too closely. The selection bias problem is structural: DoR is measured only in responders, who are identified post-randomization, making it analytically distinct from an ITT endpoint and difficult to interpret in single-arm settings where natural disease history cannot be separated from treatment effect.
The proposed remedy is RMDoR — Restricted Mean Duration of Response — a composite endpoint using multi-state modeling that incorporates both responders and non-responders into an ITT-compatible estimate. Computed as the area under the probability-of-being-in-response curve, bounded by the observation window, it is nonparametric and estimand-anchored per ICH E9(R1). The panel suggested it may be more sensitive than PFS for predicting OS in high-response-rate, cytotoxic settings — and explicitly not appropriate for cytostatic agents where response rates are low.
RMDoR is not submission-ready today; no agency has endorsed it as a primary endpoint. But with nine international regulators observing and Project SignifiCanT signaling formal guidance activity ahead, teams supporting accelerated approval programs should be watching its development. Pre-specifying it as an exploratory endpoint in an ongoing single-arm trial would be a low-cost way to generate the empirical track record it currently lacks.