For maximum assurance of conclusions, such studies need to be double-blind: neither the subjects nor the researches who interact with them are to know who's taking just chemo and who's taking the new treatment.
There's also the issue of blinding. If the control group is some outside group, then you know that everyone in the study group is receiving the new treatment. It would be much better if everyone involved didn't know whether any given patient was getting the new treatment or not until after all the results were in. For that, you need to recruit patients into the study, then randomly assign them to the new treatment or the conventional sort.
There was a lot of variance in how doctors and nurses treated the side effects, how they spoke to the patient, what other resources they provided, the quality of bedside manner, their attentiveness to lab results and patient health over time, whether they consistently washed their hands, etc.
People don't really die "of cancer" they die of things like blood loss, toxic build up from renal failure, blood clots, or infections acquired in the hospital that attack a weakened immune system, to name a few.
Doctors have tremendous latitude in how they treat patients and, as far as I know, there are no "official" guidelines for most diseases (very communicable diseases might be one major exception).
There are guidelines for developing, manufacturing, advertising, and selling drugs. However, doctors can use them off-label. From the FDA (here: http://www.fda.gov/ForPatients/Other/OffLabel/default.htm ) "From the FDA perspective, once the FDA approves a drug, healthcare providers generally may prescribe the drug for an unapproved use when they judge that it is medically appropriate for their patient. "
But my point is in response to the original parent comment. If you give treatment Y at one top hospital and your control group is everyone getting treatment X at all the other top hospitals, you still have basically anecdotal data.
If I'm very sick and the standard treatments aren't working, would I want to try a protocol that was anecdotally very successful in some clinic? Absolutely. Just because it's not scientifically proven to work doesn't mean it doesn't work.
You want to eliminate anything and everything that can (in)directly effect the outcome of treatment in a study. Ideally, the only difference at all would be the variable you are testing, but real life doesn't make things that easy.