Survival endpoints
In one sentence
A survival endpoint specifies the starting time and which events count during follow-up.
The intuition
A survival curve is a clock with a rule for stopping it. Changing either the starting point or the stop rule changes what the percentage means. “Survival” alone is not enough information to compare studies.
How it works
Overall survival (OS) counts death from any cause. Progression-free survival (PFS) typically counts progression or death. Event-free survival (EFS) uses a protocol-defined set of events; in a neoadjuvant study it can include progression preventing definitive surgery. Invasive disease-free survival (IDFS) concerns defined invasive disease events and often death after a specified starting point.
Distant recurrence-free interval and distant recurrence-free survival can handle death and other events differently. Disease-free survival also has study-specific definitions. Read the protocol and endpoint definition, rather than assuming a familiar acronym always means the same thing.
KEYNOTE-522’s co-primary endpoints were pCR (pathologic complete response) and EFS. Calling its five-year EFS an IDFS result changes the endpoint. Likewise, a postoperative biomarker cohort using a last available draw during the first year is not equivalent to a trial measuring from randomization or a uniform early postoperative draw.
An endpoint occurring sooner can make a study feasible, but it is not automatically a validated surrogate for overall survival. Regulatory approval can use different evidence depending on the indication and pathway. There is no universal requirement that a cancer drug demonstrate PFS for full approval.
Why it matters in cancer
It makes comparisons between trial results meaningful by retaining their starting points and event definitions.
Worked example
A person has a local invasive recurrence but remains alive without distant recurrence. That event may count in IDFS and not yet in a distant-recurrence endpoint; OS still has no death event. All three descriptions can be correct under their definitions.
Common confusions
- IDFS, EFS, PFS and OS cannot be relabeled to simplify a table.
- A five-year estimate needs its origin, event definition and population.
- A biomarker’s association with an endpoint is different from evidence that acting on it improves that endpoint.
Related concepts
Hazard ratio and Kaplan–Meier.
Sources and scope
General teaching, source-checked October 8, 2026. The examples are hypothetical unless a study is explicitly named. Expert and learner review remain pending.