Monday, March 07, 2005

How I would setup a medical marijuana study - I

I haven't read any of the studies, so I thought I'd create my own study using Probability and Statistics for Engineers by Miller & Freund & my knowledge of the hassles of 'subjective' studies. In this case, the response variable is level of pain from people with terminal illnesses. I'm not sure what symptoms Marijuana ameliorates for multiple sclerosis, but obviously the effects are once again highly subjective.

The #1 problem with studies such as Visual quality of a stent is how to map from an "excellent-good-fair-cruddy-miserable-crappy" scale to numbers so you can conduct a quantitative comparison. I would use a 1-10 scale. Obviously, the experiment designer can choose any size scale, e.g. 1-5, 1-100, but 10 allows sufficient granularity because you're comparing deltas. How much have my symptoms changed? Has my pain gone from a 7 to a 5? 6 to a 2? While if you get too large, say 1-100, then your patients are left wondering - Was that a 41 or a 38? on the feel bad meter.

Given that we're not measuring something exact, such as fingernail length of prisoners, then for our statistical comparison, we should use a non-parametric test - probably the U-test. Looking through my 2nd edition, the information on sample size isn't that good, so will probably have to finish this study tomorrow night with my Walpole and Myers from work. Usually non-parametric tests require a fairly sizable test group given the inexactitude of the response.

Next item is the number of groups. Ideally, you'd have a control group (placebo) and some pill form of THC and then smoking marijuana. Obviously you can't create a 'blind test' since folks will know if they're getting high, so for the sake of ease of administering, I'd just use the two groups: cannabinoids and marijuana. The other obvious risk is bias by the participants - since you can't hide the difference, you'll have to trust the integrity of your subjects.

What diseases to compare. This is hard since I'm not very knowledgeable about the symptoms and relief claimed. Knowing that pot does in fact give you the munchies, I'll choose terminal cancer victims who have a hard time eating due to stomach pain. That should also help prevent biases - folks won't fake eating/non-eating just to bias a study.

The other huge problem with clinical trials is screening out confounding variables. Are more folks using a combination of other drugs or switching drugs during the studies. Given that you can't control humans that well and their first priority is to diminish their symptoms. Would try to limit the age range & range of years that subjects have had cancer, or projected number of years until death.

Will have to continue tomorrow night, but have scanned the studies from link above and the sample sizes for the studies I see on this page range from 27 to 250. All the studies appear to be comparisons of populations: % effective with marijuana vs. % effective with cannabinoids. Does not look as though non-parametrics were used in any of the cases.

Time to body boogie & will hopefully finish up tomorrow night.

Comments:
Well thought out; great perspective on mainstream media reporting. Interesting topic, but facts get flung aside in polarized debate. It's Madness
 
Post a Comment

<< Home

This page is powered by Blogger. Isn't yours?