If your treatment works, that’s an improvement of what you had before. Once you know that, you can treat all patients. For some, that will be too late, but without your tests, it would be too late for them, too.
If your treatment doesn’t do anything at all, it keeps things the same, but the patients in the test group likely will have had some inconveniences (having to visit a doctor, getting an injection, etc), so you shouldn’t do the test.
If your treatment makes things worse, you of course shouldn’t do the test.
Problem is that you typically only can only know in hindsight which of these applies.
So, you think carefully on whether a treatment could fall in category 3, and, if so, first do it on a group of patients who consent to be Guinea pigs and, often, are already terminally ill, as any negative outcomes will cause less harm to such patients.
Then, as soon as during the test your stats tell the drug does or doesn’t work, you stop the test and either treat all patients or stop treating the test subjects.
So the comparison is between old and new, to ensure the new treatment is better than the current standard one.
Simplest "bad" reasons are the wealth of the patients. Malaria, river blindness, guinea worm, etc are terrible diseases that mostly impact poor people out of sight from Western eyes. Spending $X billion developing a drug for a population that can barely afford to feed themselves is not going to make a financial return on investment.
Best bad reason: it would cannibalize an inferior drug currently in your portfolio that's still under patent protection.
I think this behavior is at the far "evil" end of the spectrum of behaviors that drug developers systematically engage in (which I believe is far more banal + less evil than what they're accused of), but it does happen and it's a really nasty
Where it gets especially nasty is when companies buy drugs in development or pre-development from other companies in order to squash (or at least delay) a potentially competitive asset before it reaches the market.
---
Another great bad reason, but mostly applies to devices/procedures: the device/procedure is fantastic but for various structural reasons outside of the control of the device/procedure developer, there is insufficient incentive for healthcare providers to actually deploy said device/procedure.
A trivial example would be a pacemaker that requires fewer leads than the competitors and has fewer complications. Great for patients, but potentially totally uninteresting to the electrophysiologists who install it and would get paid less due to the less complex procedure.
---
Best good reasons all fall under: making a therapy is unfathomably difficult and most efforts are destined to fail OR (separately) proving a therapy works is unfathomably expensive and most efforts can't produce a positive ROI after that process.
Existing quality of treatments - if there are already efficacious drugs on the market - how sure are you that this new therapy will be best in class? Only being as good as the status quo is not an ideal competitive position. Conversely, if there is an unmet need because a disease is so lethal/debilitating, regulatory agencies can give latitude in approvals.
Likelihood patient compliance - if it is the most effective drug in the world, but requires intravenous infusion six times a day - nobody is going to adhere to that. GLP drugs are effective, but there is a needle-phobia that is preventing patients getting on board with the idea. Which is why there is an arms race for the first company to develop an oral version.
Toxicity - all chemicals are poisonous. Yet some have a lower therapeutic window than others. If you drug does what it should, but if you take 2x as much and it gives you a heart arrhythmia that is going to be a tough approval for anything but the most deadly conditions.