Homeopathic Cough Medicine

Rodney cites a 1991 analysis that does not support evidence for efficacy as evidence that homeopathy may work.
One that concluded, 15 years ago, that the current evidence was not good enough to draw a definite conclusion, but that there was a legitimate case for further research, "but only by means of well performed trials". If these "well performed trials" have been carried out, it doesn't look good for homoeopathy as more recent meta analyses such as the 2005 Lancet paper or the 2002 Paper by Edzard Ernst have tended to find no good evidence for an effect above placebo.
 
Rodney would like to position the statistics so that they show a positive outcome for homeopathy, or, at the very least, not a negative one.

M.
 
No homeopathy is a set of well defined claims.

1. If a substance causes a problem then if you dilute it by say 10x it will cause the body to remove that symptom from itself.
2.The more you dilute it the more effective it is.

As this has never been shown to be true the fundamental processes that it is supposed to work by has nothing backing it up.

Well, I was going to say basically the same as JJM. I think there is a difference between the scientific claims of homeopathy and the medical claims of homeopathy. The scientific claims are, as you say, well defined and utter nonsense. However, it could still be possible that one or more homeopathic remedies could work, either through an entirely different mechanism or through all known science being wrong.

The medical claims are a different matter though. These can be tested in the same way the JREF challenge tests things - the claim is tested with no regard to the proposed theory behind it, all that matters is that this particular remedy does something. In this case, every claim is independent. Even if a 10C preparation of Belladonna is shown to cure heart attacks, this says exactly nothing about any other remedy, or about homeopathic theory in general. This is what I meant when I said most people misunderstand the problem. I think that if any homeopathic remedy were ever shown to work, most people would then assume that homeopathy would be proven, despite the fact that this is not the case.
 
The formula that we use to find the probability of true relationships in the presence of bias is on the second page of the paper I referenced. Using the values listed above, the calculated PPV for a single study is 0.0015. That is, there is a 0.15% chance that the positive study represents a true positive. Combining these results for the 81 positive studies (these results are all independent), we have an 11% chance that there is a true effect. Linda

Using the forumula on page 2 of the paper (1st column under the section headed bias) and setting R = 0.001, beta = 0.2, u = 0.2, and assuming alpha = 0.05, I calculate the PPV for a single study of .0035 or 0.35%. I used the formula at the top of column 3 and computed a PPV of .001015 or 0.10% for 81 studies.

If you limit it to the 15 studes you say were of high quality, the PPV goes up to .00186 or a little less than .2%. This is assuming no bias, but no formulas are provided for including bias in estimated the PPV of multiple studies. This is strongly counter-intuitive in that the more studies are performed, the lower the PPV. I don't yet understand why this is true, but I'll have to mull over the math for a while before it sinks in.

For the values of R = 0.0001, beta = .5, u = .5 and alpha = .05, I get PPV values of 0.00014 for an individual study, 0.0001016 for combining 81 studies and .000186 for 15 studies.

I'm not sure about the value of R that you are using. That's a really crucial value in the formula and I don't quite follow how to estimate it with confidence. Since R represents your estimate of the ratio of true relationships to no relationships, could you explain more about why you made the choice you did for that value? Thanks.
 
Using the forumula on page 2 of the paper (1st column under the section headed bias) and setting R = 0.001, beta = 0.2, u = 0.2, and assuming alpha = 0.05, I calculate the PPV for a single study of .0035 or 0.35%. I used the formula at the top of column 3 and computed a PPV of .001015 or 0.10% for 81 studies.

If you limit it to the 15 studes you say were of high quality, the PPV goes up to .00186 or a little less than .2%. This is assuming no bias, but no formulas are provided for including bias in estimated the PPV of multiple studies. This is strongly counter-intuitive in that the more studies are performed, the lower the PPV. I don't yet understand why this is true, but I'll have to mull over the math for a while before it sinks in.

I wasn't sure about the applicability of that (multiple studies) formula to this question (I'm still mulling over it as well). If we were talking about multiple teams addressing the same question (e.g. oscillococcinum in cough), I would say yes. It may also apply to addressing the question of homeopathy, but I think one would have to present an argument to justify it. And it leads to a lower PPV. Despite Rodney's protestations, I was trying to boost the numbers by any reasonable means, so I didn't use the formula to combine the probabilities.

I think that the reason it decreases is analogous to making multiple comparisons - any "significant" differences found among multiple comparisons are more likely to be spurious than a significant difference on a single comparison.

I also realize that I deleted the values I had specified for beta and u in the first round (probably when I was revising it for clarity), so I think you used some values that were different than mine. It doesn't affect the magnitude of the outcome, though.

For the values of R = 0.0001, beta = .5, u = .5 and alpha = .05, I get PPV values of 0.00014 for an individual study, 0.0001016 for combining 81 studies and .000186 for 15 studies.

I'm not sure about the value of R that you are using. That's a really crucial value in the formula and I don't quite follow how to estimate it with confidence. Since R represents your estimate of the ratio of true relationships to no relationships, could you explain more about why you made the choice you did for that value? Thanks.

The author gave examples for R in Table 4 (e.g. epdemiology, conventional medicine trials, exploratory research (e.g. microarrays were mentioned in the text)). An R of 0.001 was suggested for exploratory research, and (if you take microarrays as an example) that is exploration that occurs within a somewhat established theoretical framework that provides some indication of constraint to your choices. At best, homeopathy could be considered equivalent to that last choice, but is more likely to be worse (there are no known methods by which homeopathy could work, so it is hard to come up with ways to constrain your choices).

I don't know if you read any more posts in this thread, but I went on to give a numerical example in #488 of one way to go about estimating true relationships and potential relationships. I can elaborate on this some more, if necessary. And I'd be interested in suggestions for other ways to estimate this parameter.

Linda
 
I wasn't sure about the applicability of that (multiple studies) formula to this question (I'm still mulling over it as well). If we were talking about multiple teams addressing the same question (e.g. oscillococcinum in cough), I would say yes. It may also apply to addressing the question of homeopathy, but I think one would have to present an argument to justify it. And it leads to a lower PPV. Despite Rodney's protestations, I was trying to boost the numbers by any reasonable means, so I didn't use the formula to combine the probabilities.

I think that the reason it decreases is analogous to making multiple comparisons - any "significant" differences found among multiple comparisons are more likely to be spurious than a significant difference on a single comparison.

I also realize that I deleted the values I had specified for beta and u in the first round (probably when I was revising it for clarity), so I think you used some values that were different than mine. It doesn't affect the magnitude of the outcome, though.

The author gave examples for R in Table 4 (e.g. epdemiology, conventional medicine trials, exploratory research (e.g. microarrays were mentioned in the text)). An R of 0.001 was suggested for exploratory research, and (if you take microarrays as an example) that is exploration that occurs within a somewhat established theoretical framework that provides some indication of constraint to your choices. At best, homeopathy could be considered equivalent to that last choice, but is more likely to be worse (there are no known methods by which homeopathy could work, so it is hard to come up with ways to constrain your choices).

I don't know if you read any more posts in this thread, but I went on to give a numerical example in #488 of one way to go about estimating true relationships and potential relationships. I can elaborate on this some more, if necessary. And I'd be interested in suggestions for other ways to estimate this parameter.

Linda

Ok. Thanks for the explanation. Post #488 was helpful. I haven't read very many of the posts in this thread - just the most recent page and then went back to your post #475 to pick up the details of the analysis to see if I could help.

BTW, the article is an interesting POV on testing. Thanks.
 
Ok. Thanks for the explanation. Post #488 was helpful. I haven't read very many of the posts in this thread - just the most recent page and then went back to your post #475 to pick up the details of the analysis to see if I could help.
Thanks for becoming involved in this thread, Beth. In addition to posts #475 and 488, I would suggest reading posts 484, 517, 518, and 522.
 
Statistics damn statistics.
And, I have a feeling they're going to damn us here. :(

However, I would like to ask Rodney if he agrees with the following excellent overview of the problem...

I think there is a difference between the scientific claims of homeopathy and the medical claims of homeopathy.

The scientific claims are, as you say, well defined and utter nonsense. However, it could still be possible that one or more homeopathic remedies could work, either through an entirely different mechanism or through all known science being wrong.

The medical claims are a different matter though. These can be tested in the same way the JREF challenge tests things - the claim is tested with no regard to the proposed theory behind it, all that matters is that this particular remedy does something. In this case, every claim is independent. Even if a 10C preparation of Belladonna is shown to cure heart attacks, this says exactly nothing about any other remedy, or about homeopathic theory in general. This is what I meant when I said most people misunderstand the problem. I think that if any homeopathic remedy were ever shown to work, most people would then assume that homeopathy would be proven, despite the fact that this is not the case.
 
Statistics damn statistics.
And, I have a feeling they're going to damn us here. :(
:)

However, I would like to ask Rodney if he agrees with the following excellent overview of the problem...
Generally, yes, and I was pleasantly surprised by Cuddles' post. He is making the same point I have been trying to make -- homeopathic medicines "can be tested in the same way the JREF challenge tests things - the claim is tested with no regard to the proposed theory behind it, all that matters is that this particular remedy does something." And that is why I strongly believe that the efficacy of homeopathic medicines should be evaluated by the same statistical model as is applied to non-homeopathic medicines.
 
Generally, yes, and I was pleasantly surprised by Cuddles' post. He is making the same point I have been trying to make -- homeopathic medicines "can be tested in the same way the JREF challenge tests things - the claim is tested with no regard to the proposed theory behind it, all that matters is that this particular remedy does something." And that is why I strongly believe that the efficacy of homeopathic medicines should be evaluated by the same statistical model as is applied to non-homeopathic medicines.
I hope you know that almost all homeopaths will run for miles from such a suggestion. They categorically refuse to be involved, and will invent the most outlandish excuses for not going this route.

The reason is quite simple: It has already been done a number of times for a number of different remedies and maladies for which they supposedly applied. Results: same as placebo in all cases. (References in this thread, and readily available across the internet from reputable sources.)

So the excuse-making is to avoid a repeat of this situation - they know all too well that ongoing patient ignorance is their bread-and-butter!
 
And that is why I strongly believe that the efficacy of homeopathic medicines should be evaluated by the same statistical model as is applied to non-homeopathic medicines.

They have been (as you well know ;)) and found not to work. This is why we get this ridiculous situation:

Science tests homepoathy in DBPC trials: results show it doesn't work.

Homeopaths test homeopathy in DBPC trials: results show it doesn't work.

Science's conclusion: homeopathy doesn't work.

Homeopaths' conclusion: homeopathy works but Double-Blind Placebo-Controlled studies are not the appropriate way to prove it!


This is the inevitable conclusion of a belief system (as that's all homeopathy is) that begins with a conclusion and distorts reality to fit.
 
The first link she provided in post 475 regarding the basic approach she used is a Bayesian analysis.

Sorry to be a bother, but would you mind explaining why this is so?

I'm trying to figure out whether my analysis of how she achieved the POV value is faulty, or whether my actual understanding of Bayesian analysis is incorrect. Or both.

Thanks!
 
Generally, yes, and I was pleasantly surprised by Cuddles' post. He is making the same point I have been trying to make -- homeopathic medicines "can be tested in the same way the JREF challenge tests things - the claim is tested with no regard to the proposed theory behind it, all that matters is that this particular remedy does something." And that is why I strongly believe that the efficacy of homeopathic medicines should be evaluated by the same statistical model as is applied to non-homeopathic medicines.

But if you agree with this, your entire argument is irrelevant. You refer to a meta-analysis of about 100 trials as basically your only evidence. Although I admit I haven't read it, I find it extremely unlikely that they all studied exactly the same remedy. Since they are all seperate claims, and you have agreed that they are, this meta-analysis is useless, since it was looking at whether homeopathy as a whole worked.

In addition, I also pointed out that even if one remedy was shown to work, this still would not provide any evidence that homeopathy is anything other than a big pile of cow poo, since it would still require all modern chemistry and biology to be wrong, all it would show is that we have one more drug whose working we do not understand yet. Given this, and given that you said you agree with this, do you now agree that homeopathy is utterly discredited?

And of course, as Zep and John have said, such studies have been done on individual remedies and have shown that they have no effect above placebo. This is why most homeopaths refuse to consider real trials and claim that homeopathy cannot be tested this way.
 
And that is why I strongly believe that the efficacy of homeopathic medicines should be evaluated by the same statistical model as is applied to non-homeopathic medicines.
OK, how about addressing the topic of this thread: homoeopathic cough medicines.

You have been provided with evidence that non homoeopathic cough medicines have some efficacy. Can you provide similar evidence for the efficacy of homoeopathic cough medicines?
 
And that is why I strongly believe that the efficacy of homeopathic medicines should be evaluated by the same statistical model as is applied to non-homeopathic medicines.

Yes! Exactly my point!

Any conventional medicine that showed up at the party as naked as a homeopathic medicine would be given the boot at the door.

Linda
 
Sorry to be a bother, but would you mind explaining why this is so?

I'm trying to figure out whether my analysis of how she achieved the POV value is faulty, or whether my actual understanding of Bayesian analysis is incorrect. Or both.

Thanks!


Without going back to her linked article, just as a general rule rule of thumb, anytime subjective assessments of probability made prior to data being collected are incorporated into an analysis, it's likely to be a Bayesian approach.
 
Yes! Exactly my point!

Any conventional medicine that showed up at the party as naked as a homeopathic medicine would be given the boot at the door.

Linda


Oh, I don't know. Depends what sort of party is is and how good-looking the medicine is naked. But it needs to be prepared to be abandoned by the next day when the party-goers are sober. :)
 
Oh, I don't know. Depends what sort of party is is and how good-looking the medicine is naked. But it needs to be prepared to be abandoned by the next day when the party-goers are sober. :)

I clearly attend the wrong sort of parties. :)

Linda
 
I clearly attend the wrong sort of parties. :)

Linda
This clearly needs to be sorted out over serious discussion. My place. Saturday night. You and Beth are invited, with medicine or without.
 
This clearly needs to be sorted out over serious discussion. My place. Saturday night. You and Beth are invited, with medicine or without.

I received my Skepdude Calendar yesterday. Which month are you? ;)

Linda
 

ISF - Join now!

Every member here is approved by hand. No bots, no spam, just people who care about evidence and honest debate.

Membership is free!

Create your free account

Back
Top Bottom