Which biases should I worry about?
[Reposted and embellished.]
>>Perhaps caution of a similiar situation would apply to Genaera as it now stands. Is this your perception, obviously taking into account that we don't have even a multiple survival bias after one trial?<<
Early in clinical development, you worry mostly about whether the observed data are fair. You want to know if the data fairly depict what the investigators actually observed (i.e. was there investigator bias?), and you want to know if the studied patient sample fairly represents the universe of patients with the specified characteristics (i.e. was there any cherry-picking of enrolled patients, what is called patient-selection bias?).
Later in clinical development, when you are satisfied that the data are fair, the important question becomes whether the data are reproducible, i.e. whether they represent a bona fide medical effect or merely a statistical fluke.
This is where program-survival bias comes into play; drugs with good data in mid-stage trials are continued and those with bad data in mid-stage trials are dropped, even though the drugs with good data may in reality be weak agents and perhaps no better than placebos.
Market Data 
Markets 
AI
