You're accidentally providing a great counter example to your own argument. Cancer is not one disease but thousands - with complex multi-factorial causes, progress and prognosis. 'Curing' cancer in one patient with a novel treatment would be an interesting data point. 'Proving' the cure was due to a given intervention (rather than say experimenter error, spontaneous or temporary remission, or the body fighting off the cancer itself) is a completely different order of problem. Generalising that treatment to other patients, testing its safety and then mass producing distributing it would be a multi-year perhaps decade long process. Inevitably, the high variance in types of cancer and responses to treatment would result in an efficacy significantly lower than 100% (I hope my understatement is clear here).
In this case ABA treatment is already well known, and there is significant experimental research into it's efficacy. It's demonstrably not effective in 'curing' austism in most people and has measured, known 'side effects' including anxiety, depression, or other mental health problems.