An issue involving independence and dependence

[Thompson,Paul]

I am involved as the statistician in a small trial of the use of a topical medication in a surgical intervention. The intervention occurs in the leg. We in the clinical research group assumed control of this trial after the PI (surgeon) started the trial. The endpoint is reoccurrence within the leg of the original condition.

 

We discovered after the fact that the PI has begun to enroll both legs in some cases, treating each separately. The PI wishes to treat each leg independently. I do not. In addition, at the end of the trial, an unknown proportion of cases will  have two legs treated – to date, it is 1 of 17 cases or thereabouts. Of the 50 cases who are to be treated, we could possibly expect 3-5 to have 2 legs treated.

 

I have contended that any publication of this trial will require evidence that the assumption of independence of the two limbs for a given person is a reasonable one. The PI wishes to assume independence simply on his say-so.

 

For those of you who have and are currently serving as referees, what say you? Must a strong case for independence be made? Or would you allow a trial with a simple assumption of independence to be considered?

 

Paul A. Thompson

 

 

----------------------------------------------------------------------- Confidentiality Notice: This e-mail message, including any attachments, is for the sole use of the intended recipient(s) and may contain privileged and confidential information. Any unauthorized review, use, disclosure or distribution is prohibited. If you are not the intended recipient, please contact the sender by reply e-mail and destroy all copies of the original message.

[Marc Schwartz]

On Jul 10, 2012, at 1:53 PM, Thompson,Paul wrote:

> I am involved as the statistician in a small trial of the use of a topical medication in a surgical intervention. The intervention occurs in the leg. We in the clinical research group assumed control of this trial after the PI (surgeon) started the trial. The endpoint is reoccurrence within the leg of the original condition.
>
> We discovered after the fact that the PI has begun to enroll both legs in some cases, treating each separately. The PI wishes to treat each leg independently. I do not. In addition, at the end of the trial, an unknown proportion of cases will have two legs treated – to date, it is 1 of 17 cases or thereabouts. Of the 50 cases who are to be treated, we could possibly expect 3-5 to have 2 legs treated.
>
> I have contended that any publication of this trial will require evidence that the assumption of independence of the two limbs for a given person is a reasonable one. The PI wishes to assume independence simply on his say-so.
>
> For those of you who have and are currently serving as referees, what say you? Must a strong case for independence be made? Or would you allow a trial with a simple assumption of independence to be considered?
>

Paul,

Is this situation the same one that you posted on back in April?

https://groups.google.com/d/topic/medstats/ltydfqDfTE4/discussion

If not, then perhaps the prior discussion might be apropos here and Prof. Altman's references in his reply in that thread might be of benefit.

Regards,

Marc Schwartz

[M Behnke]

I would definitely not treat the legs as independent, because response to treatment is very much dependent on patient factors such as immune status etc that will affect both legs in the same way.  

You could test this by calculating the correlation  in response between right and left legs, but if (as I suspect) correlation is high, you're stuck with deciding how to appropriately include their data (which is NOT to choose the leg that had the response the PI wants ;)

Mik

--
To post a new thread to MedStats, send email to MedS...@googlegroups.com .
MedStats' home page is http://groups.google.com/group/MedStats .
Rules: http://groups.google.com/group/MedStats/web/medstats-rules

[Peter Flom]

On Jul 10, 2012, at 1:53 PM, Thompson,Paul wrote:

> I am involved as the statistician in a small trial of the use of a topical medication in a surgical intervention. The intervention occurs in the leg. We in the clinical research group assumed control of this trial after the PI (surgeon) started the trial. The endpoint is reoccurrence within the leg of the original condition.
>
> We discovered after the fact that the PI has begun to enroll both legs in some cases, treating each separately. The PI wishes to treat each leg independently. I do not. In addition, at the end of the trial, an unknown proportion of cases will  have two legs treated – to date, it is 1 of 17 cases or thereabouts. Of the 50 cases who are to be treated, we could possibly expect 3-5 to have 2 legs treated.
>
> I have contended that any publication of this trial will require evidence that the assumption of independence of the two limbs for a given person is a reasonable one. The PI wishes to assume independence simply on his say-so.
>
> For those of you who have and are currently serving as referees, what say you? Must a strong case for independence be made? Or would you allow a trial with a simple assumption of independence to be considered?
>

I serve as a statistical referee for PLoS Medicine, and not only would I NOT accept an author’s “say-so” that two legs on one person were independent, but I would be extremely skeptical of any evidence that they were. If a test said my right leg was unrelated to my left, I would throw out the test.  If reading such an article, I would probably suggest using only one leg from each person.

Peter

[SR Millis]

Or use an analytic approach to deal with the dependency in the observations, eg, GEE.

SR Millis

---

From: Peter Flom <peterflom...@mindspring.com>
To: meds...@googlegroups.com
Sent: Wednesday, July 11, 2012 2:46 PM
Subject: RE: {MEDSTATS} An issue involving independence and dependence

--
To post a new thread to MedStats, send email to MedStats@googlegroups.com .

[Peter Flom]

In general, I’d agree with this suggestion, but with so few data being paired, I think it better to KISS and lose a small amount of data rather than go complex for a slightly larger N

 

Peter

--
To post a new thread to MedStats, send email to MedS...@googlegroups.com .

MedStats' home page is http://groups.google.com/group/MedStats .
Rules: http://groups.google.com/group/MedStats/web/medstats-rules

 

--
To post a new thread to MedStats, send email to MedS...@googlegroups.com .