Correlation is not a causation.

If you eat a sugar pill with an "imprint of camomile" and it makes you feel different, it's neither the sugar nor the imprint, but your own belief that does that. However the "evidence based medicine" may conclude it was the spirit of camomile or the lovely sperical shape that made all the difference. Evidence must be supplanted by the reliable reproduction of the results with all variable factors accounted for. Otherwise it's just guessing.

evidence: cutting off his leg is bad, but we had to do it first to have the evidence. => data

science: from all we know, we assume that cutting off his leg is bad, so we don't do it. => data and models

A lot of "alternative medicine" is just not science, and simply does not work. Unless you consider being temporarily "cured" by your own endorphins as an instance of evidence based medicine.