I disagree with "impossible." The ABBA test is a very good beginning. There of course might be some confounding influences, but one way to deal with these is to do a series of ABBA tests of varying lengths, and also a series at different dosages. A very strong effect may show up across different designs. If it doesn't, well, you could have missed it, sure, but you are probably more interested in strong effects than weak effects, so if you show no effect across different designs you might accept "no" for an answer. If it shows up in just one design, then you have a clue to follow.