[gdpawel]

In clinical trials, an event or outcome can be measured objectively to determine whether the intervention being studied is beneficial. The endpoints of a clinical trial are usually included in the study objectives. Some examples of endpoints are survival, improvements in quality of life, relief of symptoms, and disappearance of the tumor.

Changing from pre-specified endpoints once a trial has begun can introduce bias. If endpoints of interest are changed to get better results, the clinical trial will produce biased results which do not inform research. Any changes in endpoints should be declared and explained to the registry of the clinical trial and to any journals that manuscripts are submitted to.

A well designed randomized controlled trial should be in a state of what ethicists call clinical equipoise: uncertainty in the expert community as to the comparative merits of drugs in either arm of the trial (if it weren't in equipoise, the study would be unethical in part because it would knowingly subject some patients with a life-threatening illness to an intervention thought to be inferior).

If researchers are running trials exactly as they are supposed to, 50% of trials will give an inconclusive result in which neither arm of the trial is superior. In another 25% of trials, the "new" drug should prove superior, and in the remaining 25% of the trials, the "old" drug should prove better.

This pattern has been roughly (and encouragingly) established for NCI funded RCTs in a series of papers published by Benjamin Djulbegovic (the latest of which was just published in the March 24 2008 issue of Archives if Internal Medicine).

The field of cancer research has done a lousy job picking candidates; the rate of success going from phase 1 through to registration is around 5%. The big problem, some say, has to do with poor quality animal studies and phase 1 trials.