Every so often I have a "moment of inspiration". Here is my last one! :-) It's on sub-types in stuttering, and how the power of brain experiments could be improved using knowledge from empirical pharmacology.
1) It is not clear whether there are sub-types or not. However, if sub-types exist, they will significantly complicate any experiment. I mentioned the issue of false negative in my last post, where a compound might not be effective enough in a random control trial because only a minority significantly benefits and the majority not at all. There could be two sub-types, A and B, and the compound is only effective for type A. Or, let's imagine brain scans. And let's assume stuttering is caused by an underactivation of region A or by an underactivation of region B. Of course, in brain research, controls are compared to a mixture of type A and B stutterers. The signal analysis would show an underactivation in both A and B but with only half the strength, and a statistical test might show not statistically significant signals. To conclude, sub-types significantly dampen signals. The same phenomenon could occur in stuttering therapy evaluations.
2) There is somewhat evidence for sub-types. People who stutter seem to react differently to different medication, or drugs like alcohol, dope, or ecstacy. Also, some people seem to react much better to speech therapy that others.
3) There are also theoretical considerations. Speech is a very complicated multi-step and parallel process. However, the effect that characterises stuttering is simple, the absence of an appropriate signal. So many different malfunctions in this complicated process could go wrong, which would lead to the same symptom, the absence of the appropriate signal. Imagine a news speaker reading from a prompter. She will stutter if the prompter gets stuck. But this getting stuck could have several different causes: computer crash or slow-down, the editor did a mistake, and so on. Viewed from this perspective, it would be very surprising if there were no sub-types!
4) There is a chicken and egg problem, because we cannot know whether sub-types exist and identify them because our experiments do not have enough power because we do not separate the sample into different sub-groups!
Here is my solution:
a) put people who stutter in different group according to their reaction to medication or drugs.
b) do brain scans by comparing a sub-group to controls.
This will allow to get much better signals. So again, use empirical pharmacology to form sub-groups and analyse them.
1 comment:
So selection bias isn't so bad after all. Subjects who have no perceived improvement will tend to drop out of the trials ar greater rates than those who do improve. The researchers can then further reduce the sample by culling out those who stay with the trial but show no improvement. Then those subjects remaining after this process fall into two groups: the subjects who are genuinely affected by the treatment and those who respond to the treatment as a placebo effect (i.e., they would have responded just as well if they were given a placebo!)
Post a Comment