This topic is pretty dead, but I'll try to clarify some last points:
The birth control example was meant to show that scientists and doctors need to consider the actual statistical outcomes of different treatments and not the ideal outcomes. The treatment option in that set with the best "real" outcome also had the worst ideal outcome, but for modern medicine the "real" results are what matter.
You never provided a reference for the study you first brought up, so arguing about it isn't really going to be productive. My guess was that the study you were mentioning would have included control groups where they informed a similar sets of people about AA, control non-AA program, a unrelated control program (e.g. a writing club), or did not inform them of any program. If this were the case, then any significant differences should be due to the program they were informed about (they should all have similar stats on people not showing up at all - the 30% you mentioned). The study you mentioned could have been a flawed study that overstated its conclusions or it could have been a perfect study, but you seem to be the only person in this discussion that has knowledge of it.
Without a well controlled (demographic or experimental) scientific study, it should not be assumed that AA is the best course of treatment for a patient.
Modern medicine should not give up when a patient is non-compliant just because it is unethical to force them into treatment. Improved treatments and ways of informing patients about them (in order to get their consent or compliance) is needed when there are problems such as this.
The AC probably meant that if you only count the successes, then your success rate will be 100%. If you exclude a set of data from one condition then you have to exclude it from the controls as well (so you can properly compare them). If all data is included when comparing different conditions, then other variables that are independent of the conditions should fall to background noise (e.g. patients dying in a plane crash or not showing up to any meeting should be the same for both conditions and not affect the end conclusion)
If you are unable to force a patient to go to a program, then the beginning of the treatment starts when you inform them and encourage them to go. A large problem with treatments starting this way is that many patients will not listen. In this case the treatment failed for the patients that did not listen.