Dear Ben,

 

In my opinion the philsophers are deeply confused about this.

 

It is not necessary for the purpose of issuing a probability statement to know the  value of  unmeasured covariates; it is sufficient to know their distribution in probability.

 

The following analogy may be helpful. Suppose we have to calculate the probability that sum of scores  on two dice, a red die and a black die is 10. Actually one of the dice has alread been rolled, the red die, but we have not seen the score. It is sufficient, however, that we know that a necessary condition that the total score be 10 is that the score for the red die be 4,5, or 6. We know that this probability is 1/2 and we know that we now require a 6.5,or 4 on the black dies as the case may be and that in each case this probabability is 1/6. Hence the probability that the total a score is 10 is 1/12.

 

Of course if somebody now reveals the score on the red die to us we can refine our probability assessment of 1/12. If the score is 1,2,or 3 the  probability is now 0 since no value of the black die can make the total 10. if the score is 4,5,6 the probability is now 1/6. In fact the probability of 1/12 that we had before can be seen as being composed of (1/2 x 0) + (1/2 x 1/6).

 

Similarly the fact that if confounders were revealed we would modify the probability assessment we have from an RCT does not mean that the probability assessment is invalid. It is a valid statement given what we know at that stage. All probability statements are conditional on what is known. The fact that if we knew more we would refine them does not make them unvalid when we don't know more.

 

This is so obvious that I must confess to be underwhelmed by the level of understanding of this by certain critics of the RCT.

 

See

 

Senn, SJ. Baseline Balance and Valid Statistical Analyses: Common Misunderstandings, Applied Clinical Trials 2005; 14: 24-27.

 

For a discussion.

 

Stephen

 

 

to add to Carolin's excellent post, Ian, can we also speculate that maybe besides randomization to equate the confounding factors across trial arms, that maybe once we ensure a large enough sample size, we will mitigate the role of confounding factors and ensure that the known and unknown factors are evenly spread out. In short, maybe one can argue that even if well designed and conducted, a RCT with small sample size arms can run the risk of an uneven distribution of factors. this issue has always intrigued me...can you comment? 

 

I still refer to Bradford-Hill 9 when interpreting causation....
 

 

 

 

 

 

Best,

 

Paul E. Alexander

 

 

 

 

 

 

---

From: Ian Johnson (CMA) <[log in to unmask]>
To: [log in to unmask]
Sent: Fri, January 28, 2011 10:34:23 AM
Subject: Re: Can RCT help establish causation?

Indeed, the fundamental principle in order to ensure that the treatment arm demographics are as similar as possible. Peto tried to refine this in the ISIS trials, where patients entering the study were assigned to a treatment arm based on their characteristics to predefine identical populations in each treatment arm, and spawned the observed vs expected statistic. This is difficult to achieve in an observational study, unless it is extremely large, and has sufficient power to segregate the confounding factors, so RCTs are likely to remain the method of choice. It is the discussion around the comparator treatment and the disconnect between the FDA and EMA requirements that is the cause of much debate.

 

Ian

 

From: Evidence based health (EBH) [mailto:[log in to unmask]] On Behalf Of Caroline Boulind
Sent: 28 January 2011 15:19
To: [log in to unmask]
Subject: Re: Can RCT help establish causation?

 

As far as I understand it, that is the purpose of randomisation. If it is done correctly, randomisation distributes all varying factors among the test population evenly between the arms of the study. Obviously, one can't identify all factors that might possibly be related to the intervention and outcome of interest to you, so the important thing about randomisation is that both known and unknown 'confounders' are (theoretically) evenly distributed. This means that the only thing changing between the arms of a trial is the intervention of interest, and differences in outcome between the two arms can therefore be interpreted as being a consequence of the intervention of interest.

I suppose the problem is that there is no way of identifying those so called unknown confounders and it is therefore impossible to say definitively that there are no alternative causative variables that are differentially distributed between the arms of the study...

 

For me the important point is that a well conducted and reported RCT is infinitely closer to being able to identify a causal relationship between exposure and outcome than an observational study. Until a better design is identified EBM is better placed taking its lead from well conducted RCTs than observational studies alone.
Having said this, I think there is a place for observational studies to be considered in conjunction with RCTs, especially in areas of medicine (such as surgery) where RCTs may be more difficult to implement well.

 

Carolin

 

Dr. Caroline Boulind
Clinical Research Fellow
01935 384559e
>>> "Djulbegovic, Benjamin" <[log in to unmask]> 28/01/2011 15:06 >>>

Dear all

I'd like to post this question to the group that I have been thinking about for some time... Is there a scientific method that allows us to LOGICALLY  distinguish the cause-effect from the coincidence? David Hume, one of the most influential philosophers of all times, concluded that there is no such a  method. This was before RCTs were "invented". Many people have made cogent arguments that (a well done) RCT is the ONLY method that can allow us to draw the inferences about causation. Because this is not possible in the observational studies, RCTs are considered (all other things being equal) to provide more credible evidence than non-RCTs. However, some philosophers have challenged this supposedly unique feature of RCT- they claim that RCTs  cannot (on theoretical and logical ground) establish the relationship between the cause and effect any better than non-RCTs. I would appreciate some thoughts from the group:

1. Can RCT distinguish between the cause and effect vs. coincidences? (under which -theoretical- conditions?) 

If the answer is "no", is there any other method that can help establish the cause and effect relationship?

I believe the answer to this question is of profound relevance to EBM.

 

Thanks 

 Ben Djulbegovic 

 

 

This e-mail and any attachments may contain confidential and
privileged information. If you are not the intended recipient,
please notify the sender immediately by return e-mail, delete this
e-mail and destroy any copies. Any dissemination or use of this
information by a person other than the intended recipient is
unauthorized and may be illegal.

---

This email may contain confidential or legally privileged information, intended only for the addressee.
If you have received this email in error, you are hereby notified that any disclosure, copying, distribution or reliance upon the contents of this email is strictly prohibited. Please contact the sender to arrange for correct delivery, and then delete this email.

Any views or opinions presented in this email are those of the author. Please check the email for viruses, as the sender accepts no liability for damage caused by a virus transmitted via this email.

Registered Office: Complete House, 19-21 King Edward Street, Macclesfield, Cheshire SK10 1AQ, UK
Tel: +44 (0)1625 624000, Fax: +44 (0)1625 619812, Email: [log in to unmask], Web: www.complete-grp.com
Registered in England No. 2503062