Test-retest (continuous variable)

[John B]

Hi,
I want to classiffy whether patients changed blood glucose before and
after a therapy was given. The idea is to classify patients in 2
groups: Group 1: no reduction in glucose after treatment, Group 2:
patients who experienced a significant change in glucose after
treatment. The problem I face is that test-retest for the glucose
assay we use has a variation coefficient of 6% (figure obtained from
the manufacturer). How this problem is usually solved?

Would it be correct to consider changes of +/- 6% as "no changes" and
therefore consider only as "responders" just those patients in whom
glucose levels were above or below 6%?

Any suggestion or link would be grearly appreciated.

John

[BXC (Bendix Carstensen)]

It seems a bad idea to throw away data by dichotomizing your response --- any particular reason for not analysing the continuous outcome?
Bendix Carstensen
_______________________________________________

Bendix Carstensen
Senior Statistician
Steno Diabetes Center
Niels Steensens Vej 2-4
DK-2820 Gentofte
Denmark
+45 44 43 87 38 (direct)
+45 30 75 87 38 (mobile)
b...@steno.dk http://www.biostat.ku.dk/~bxc
www.steno.dk

[John B]

Bendix,

Thanks for your reply. The experimental model is taking basal samples
for glucose giving the treatment and to extract new samples after two
hours. A previous study shows a significant reduction in glucose. So,
this is a before-after design with a continous variable measured at
two single points.

[SR Millis]

Wouldn't you consider using an analysis of covariance using the baseline measure as the covariate and the post-treatment continuous variable as the response variable?

Scott Millis


--- On Mon, 12/7/09, John B <bangali...@gmail.com> wrote:

[John Uebersax]

Hello John,

> The problem I face is that test-retest for the glucose
> assay we use has a variation coefficient of 6%

I haven't seen the term "variation coefficient" used in this context
before. Does this mean the standard error of the difference, a
coefficient of variation, or something else?

> Would it be correct to consider changes of +/- 6% as "no changes"

An issue with that is why choose +/- 6%, as opposed to, say, +/- 1.96
* 5% (1.96 being the critical z value of a confidence interval).

The literature on psychometric measurement theory abounds with
suggestions about how to make practical use of knowledge of test-
retest reliability. For example, from a test-retest correlation
coefficient one can predict the reliability of an average of repeated
measurements. In theory one could take the measurement 2 or more
times and achieve any set degree of reliability that way.

John Uebersax

[John B]

Scott,
Never thought for analysis of covariance for this tye of measurements
probably becasue of my ignorance. Can you suggest any reading?
Thanks

[John B]

John,

Thanks for your reply.

My problem is to separate real changes from changes due to the
variability- imprecission of the assay.

>I haven't seen the term "variation coefficient" used in this context
> before.  Does this mean the standard error of the difference, a
> coefficient of variation, or something else?

The coefficient of variation is used as a measure of precission in
many assays. Another terminology is Between-run precision (copied
form Google Books): Is an Index of the ability of
the assay to reproduce the same result on the same sample from run to
run and from day to day. Where samples are routinely run as singlen
determinations the between-run precision is due to a combination of
the errors within and between assays.

My understanding is that CV is the ratio of the standard deviation to
the mean. The coefficient of variation describes the magnitude sample
values and the variation within them. For this particular test
(glucose) the CV is 6%.

> > Would it be correct to consider changes of +/- 6% as "no changes"
> An issue with that is why choose +/- 6%, as opposed to, say,  +/- 1.96
> * 5% (1.96 being the critical z value of a confidence interval).

Becasue this is the only information we hae regarding the test-retest
reliability of this measurement. I copy the info I got from the
manufacturer:
Typical performance of the analytical methods
Within Run RSD 4.0 %
Between Run RSD 5.7%

Many thanks for your help

John B.

[Bruce Weaver]

On Dec 8, 3:37 am, John B <bangali.doc...@gmail.com> wrote:
> Scott,
> Never thought for analysis of covariance for this tye of measurements
> probably becasue of my ignorance. Can you suggest any reading?
> Thanks

You could take a look at:

http://www.bmj.com/cgi/reprint/323/7321/1123.pdf

--
Bruce Weaver
bwe...@lakeheadu.ca
http://sites.google.com/a/lakeheadu.ca/bweaver/Home
"When all else fails, RTFM."

[John B]

Bruce,

Thanks a lot for you reference. I have read it. Actually the baseline
glucose and postreatment glucose are highly correlated (0.89). If I
have to use ANCOVA, the dependend variable should be post-treatment
glucose. In SPSS this analysis asks for Fixed factor and covariates. I
guess covariate should be baseline glucose. What about fixed factor?

Thanks

On Dec 8, 1:21 pm, Bruce Weaver <bwea...@lakeheadu.ca> wrote:


> You could take a look at:
>
>  http://www.bmj.com/cgi/reprint/323/7321/1123.pdf
>
> --
> Bruce Weaver

> bwea...@lakeheadu.cahttp://sites.google.com/a/lakeheadu.ca/bweaver/Home

[SR Millis]

John,

Fixed factor is your group (ie, treatment vs placebo).



~~~~~~~~~~~
Scott R Millis, PhD, ABPP (CN,CL,RP), CStat, CSci
Professor & Director of Research
Dept of Physical Medicine & Rehabilitation
Dept of Emergency Medicine
Wayne State University School of Medicine
261 Mack Blvd
Detroit, MI 48201
Email: aa3...@wayne.edu
Email: srmi...@yahoo.com
Tel: 313-993-8085
Fax: 313-966-7682

--- On Tue, 12/8/09, John B <bangali...@gmail.com> wrote:

> From: John B <bangali...@gmail.com>
> Subject: {MEDSTATS} Re: Test-retest (continuous variable)
> To: "MedStats" <meds...@googlegroups.com>

[SR Millis]

John,

I addition to the excellent paper already suggested, I've found this book helpful:

Bonate, P. (2000). Analysis of pretest-posttest designs. Boca Raton, FL: Chapman & Hall / CRC.

~~~~~~~~~~~
Scott R Millis, PhD, ABPP (CN,CL,RP), CStat, CSci
Professor & Director of Research
Dept of Physical Medicine & Rehabilitation
Dept of Emergency Medicine
Wayne State University School of Medicine
261 Mack Blvd
Detroit, MI 48201
Email: aa3...@wayne.edu
Email: srmi...@yahoo.com
Tel: 313-993-8085
Fax: 313-966-7682


--- On Tue, 12/8/09, John B <bangali...@gmail.com> wrote:

> From: John B <bangali...@gmail.com>
> Subject: {MEDSTATS} Re: Test-retest (continuous variable)
> To: "MedStats" <meds...@googlegroups.com>

[John B]

Johm,

Thanks. There is no placebo here only treatment. Before and after
treatment.

On Dec 8, 7:31 pm, SR Millis <srmil...@yahoo.com> wrote:
> John,
>
> Fixed factor is your group (ie, treatment vs placebo).  
>
> ~~~~~~~~~~~
> Scott R Millis, PhD, ABPP (CN,CL,RP), CStat, CSci
> Professor & Director of Research
> Dept of Physical Medicine & Rehabilitation
> Dept of Emergency Medicine
> Wayne State University School of Medicine
> 261 Mack Blvd
> Detroit, MI 48201
> Email:  aa3...@wayne.edu

> Email:  srmil...@yahoo.com
> Tel: 313-993-8085
> Fax: 313-966-7682

[John B]

Thanks. I got it. There is a full chapter on ANCOVA.

[SR Millis]

John,

I'm sorry but I misunderstood your design. If you have only a single group, you're pretty much limited to a paired t-test or the Wilcoxon signed rank test. Of course, given your experimental design, you cannot really make any inferences regarding the efficacy (or lack thereof) of your treatment.

Scott Millis

--- On Tue, 12/8/09, John B <bangali...@gmail.com> wrote:

> From: John B <bangali...@gmail.com>
> Subject: {MEDSTATS} Re: Test-retest (continuous variable)
> To: "MedStats" <meds...@googlegroups.com>

[John B]

Scott,

Anyhow, thanks for your input

On 8 Dec, 22:10, SR Millis <srmil...@yahoo.com> wrote:
> John,
>