Progression-free survival

Progression-free survival (PFS) is the time from the start of treatment until disease progresses or the patient dies of any cause, whichever comes first. It differs from overall survival (OS) in counting progression as an event, which makes it reachable far sooner and with fewer patients — the reason it has become a common primary endpoint in oncology trials and a frequent basis for regulatory approval, particularly where long post-progression survival would make an OS trial impractically large or slow. It is assessed by standardized radiologic criteria, most often RECIST, and is reported alongside objective response rate and duration of response.

Its weaknesses are consequences of how it is measured. Progression is detected only when a scan is performed, so PFS is interval-censored and its apparent value depends on the imaging schedule; unequal or unblinded assessment between arms biases the result, which is why blinded independent central review is used to check investigator assessment. Patients who stop treatment for toxicity and are no longer scanned may be censored informatively, in a manner correlated with the outcome being measured. And because death from any cause counts as an event, PFS counts mortality as an event, though it is generally unaffected by therapies given after progression.

The deeper question is what it stands for. PFS is valuable as a direct measure of disease control and of the interval before a patient's cancer worsens, but its correlation with overall survival is inconsistent across tumor types and drug classes, and a statistically significant PFS gain of a few weeks may not correspond to any survival or symptom benefit. This gap is sharpest with checkpoint inhibitor regimens, where delayed separation of curves and pseudo-progression complicate the endpoint further. PFS remains useful for comparing treatment sequencing strategies and identifying prognostic factors, and in lung cancer, renal cell carcinoma, multiple myeloma and other solid tumors it is routinely reported — but as a surrogate for survival it requires validation within each specific setting rather than assumption.