Overall survival is the time from randomization into a treatment until patient death. This is generally regarded as the “gold standard” for oncology endpoints. It is a very powerful data point, because it takes into account deaths from cancer and those from toxicity. So if you have a highly effective cancer killer, but it’s fatally toxic, then the median OS will show you that the drug is not a suitable treatment option.
Unfortunately, OS as an endpoint can run into a number of conflicts. For starters, patients may live for a long time with certain forms of cancer, like early-stage breast cancer or CML. Patients could survive for decades with these cancer types. Thus, overall survival is a very challenging endpoint to use in those situations. Second, as the number of treatment options in a tumor area improve, we run into the challenge of interpreting OS. If a patient progresses while on placebo in a study for drug X, many of them will seek other options. So the patient may then proceed to a new trial for drug Y and extend his or her survival. This is great news for the patient. But it can ruin the data interpretation for drug X, because the placebo looks like it helped the patient, too.
Despite these challenges, the FDA typically requires confirmation of an overall survival benefit, even if response rates or progression-free survival led to initial approval.