Optimal Data Analysis

[John Norton]

Hi List,
 
What is the best way to determine the optimal data value from a scale level variable and relative to a binary outcome variable?  Specifically what I'm looking for here is a value or cut point on the scale variable above which we start to see an increase in the incidence of X.  An example might be the optimal psi reading at which artificial corneas burst. Or, what is the optimal proportion of the pelvis irradiated at 20 Gy above which we start to see an increase in the incidence of toxicities?
 
I know there is a statistical application available (http://www.apa.org/software/odt/) but I was hoping SPSS could do this.  Can anyone offer advise?
 
Thanks,
 
 
 
John Norton
Biostatistician
Oncology Institute
Loyola University Medical Center
 
[phone]
[hidden email]
 
"Absence of evidence
      is not evidence of absence"

[Marta García-Granero]

Hi John

You know, what you describe looks a lot like ROC analysis. Why do you
want to use a different tool? What is the advantage of using that ODA
methodology?

Monday, September 18, 2006, 7:32:58 PM, You wrote:

JN> What is the best way to determine the optimal data value from
JN> a scale level variable and relative to a binary outcome variable?
JN> Specifically what I'm looking for here is a value or cut point on
JN> the scale variable above which we start to see an increase in the
JN> incidence of X.  An example might be the optimal psi reading at
JN> which artificial corneas burst. Or, what is the optimal proportion
JN> of the pelvis irradiated at 20 Gy above which we start to see an
JN> increase in the incidence of toxicities?

JN> I know there is a statistical application available
JN> (http://www.apa.org/software/odt/) but I was hoping SPSS could do
JN> this.  Can anyone offer advise?


JN> "Absence of evidence
JN>       is not evidence of absence"

(Douglas G Altman and Martin J Bland, BMJ 1995;311:485 & BMJ
2004;328:476-477, I like those two papers a lot, and I cite them
constantly too)




--
Regards,
Dr. Marta García-Granero,PhD           mailto:[hidden email]
Statistician

---
"It is unwise to use a statistical procedure whose use one does
not understand. SPSS syntax guide cannot supply this knowledge, and it
is certainly no substitute for the basic understanding of statistics
and statistical thinking that is essential for the wise choice of
methods and the correct interpretation of their results".

(Adapted from WinPepi manual - I'm sure Joe Abrahmson will not mind)

[Anthony Babinec]

John,
The ODA package by Yarnold and Soltysik implements
"optimal discriminant analysis." This finds a cutpoint
that minimizes errors in assignment to the groups
via a linear programming approach. Some ways for
you to find this sort of cutpoint might include:
discriminant analysis, logistic regression, or a
CART-like procedure that does binary splits. Whether
any of these do what you want - find "a value or cut point
on the scale variable above which we start to see an
increase in the incidence of X" - isn't clear to me.

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
John Norton
Sent: Monday, September 18, 2006 12:33 PM
To: [hidden email]
Subject: Optimal Data Analysis

Hi List,

What is the best way to determine the optimal data value from a scale level
variable and relative to a binary outcome variable?  Specifically what I'm
looking for here is a value or cut point on the scale variable above which
we start to see an increase in the incidence of X.  An example might be the
optimal psi reading at which artificial corneas burst. Or, what is the
optimal proportion of the pelvis irradiated at 20 Gy above which we start to
see an increase in the incidence of toxicities?

I know there is a statistical application available
(http://www.apa.org/software/odt/) but I was hoping SPSS could do this.  Can
anyone offer advise?

Thanks,



John Norton
Biostatistician
Oncology Institute
Loyola University Medical Center

[phone]
[hidden email]

"Absence of evidence
      is not evidence of absence"