Evidence and alternative medicine.

The most important thing we have to decide about alternative therapies is whether to test them in a clinical trial format. In order to decide this, we need to compare the prior probabilities of alternative therapy hypotheses with the prior probabilities of conventional therapy hypotheses that end up in clinical trials. If the prior probabilities of the alternative therapy hypotheses are lower than the prior probabilities for their competing conventional hypotheses, we do not have a good reason to test them. There are basic science reasons, source reasons, and methodologic reasons for assigning low prior probabilities to typical alternative therapy hypotheses--indeed, hypotheses that the NIH Office of Alternative Medicine supports. Basic science reasons are of two types: whether the hypothesis agrees with a piece of basic science that does not have a good deal of support, and whether the hypothesis is logically incompatible with a well-supported piece of basic science. The source reasons are about the educational and other characteristics of the people who discover, pursue, and promote a therapy hypothesis. The methodologic reasons are about how well a therapy hypothesis satisfies methodologic criteria such as simplicity and modesty. Other things being equal, a hypothesis that is simpler and more modest than other is more plausible as well. Problems arise also about using positive cases in assigning a prior probability to an alternative-therapy hypothesis because virtually every alternative therapy has positive cases, but not every hypothesis is true. The NCI and OTA "best-case" review approach does not solve this problem because it does not include a rule for distinguishing (in a reliable way) between positive cases that show something about a therapy hypothesis being true and positive cases that show nothing of the kind. This paper (as long as it is) is no more than an introductory discussion of some basic points involved in making sensible estimates of the prior probabilities of hypotheses.(ABSTRACT TRUNCATED AT 400 WORDS)