We can always use more case studies of insanity that aren't religion, right?
Well, Miracle Mineral Supplement is my new go-to example for Bad Things happening to people with low epistemic standards. "MMS" is a supposed cure for everything ranging from the common cold to HIV to cancer. I just saw it recommended in another Facebook thread to someone who was worried about malaria symptoms.
It's industrial-strength bleach. Literally just bleach. Usually drunk, sometimes injected, and yes, it often kills you. It is every bit as bad as it sounds if not worse.
This is beyond Poe's Law. Medieval blood draining via leeches was far more of an excusable error than this, they had far less evidence it was a bad idea. I think if I was trying to guess what was the dumbest alternative medicine on the planet, I still would not have guessed this low. My brain is still not pessimistic enough about human stupidity.
Historically, most drugs have been identified by high-throughput screening, i.e. you purify an enzyme of interest and test billions of different chemicals against it for the desired effect. You then test for an effect in cell culture (compared to healthy cells), or you can screen directly against the cancer cells. Once you have that evidence, you test whether it has effects in mice, and only after that can you test anything in humans.
It's possible to propose a single chemical and get it right by chance, but testing a single chemical is cheap. In an already-equipped lab, the initial cell culture data will probably take a few weeks and under a thousand dollars, and after that you will have people willing to help and/or fund you. The lack of even this initial evidence is generally a good reason to believe that something doesn't work.
With regards to hypotheses, a lot of the early drugs were identified by chance - there's a description at History of cancer chemotherapy. Most of the current interest is in targeted therapy, i.e. intended to act against specific proteins involved in various types of cancer, and the starting point is the identification of that protein. Chemo drugs are a bit different since they're a very broad class (they target rapidly dividing cells in general, which is also what causes the toxicity), and the metabolic networks they affect are generally well-known, so the initial hypotheses tend to be about new ways that you can intervene in those networks. There are other approaches to the various steps as well, e.g. structure-based drug design has had some success, but not yet enough to replace the screens.