Re: {MEDSTATS} Re: Comparison between two different grading scale

[Peter Flom]

Heart <nicma...@gmail.com> wrote
>
>I'll try to explain better my request.
>The two scales work for both disease. The diseases are TIA (transient
>ischemic attack, less severe cerebral disease) and stroke (more severe
>disease), so the ratio of prevalences in every group gives
>informations on the severity of cerebral ischemic disease.
>n1 and n2 are the prevalences of TIA (n1) and stroke (n2).
>My question: is there a statistical methods that describes the
>differences between the two grading scales? The best grading scale is
>the one that describes a proportionality of disease, in my opinion,
>but i don't know the statistical method to use
>

OK, let's see:
You do the test with method 1 and it says .... 1 through 5.
You do the test with method 2 and it says .... 1 through 5.

Now, do you decide, based on these tests:
No, disease, TIA, stroke?

Or do you know that everyone has one or the other, and thus decide
TIA, Stroke?

Or do you do one test for TIA and another for stroke?

If it's first how do you decide on two cutoff scores?
If it's the second or third, how do you decide on a single cutoff?
If it's the third, how do you combine the two diseases?

There will be methods to describe the differences ... you could test difference of proportions,
probably. But both methods will describe the 'proportionality of disease', and which does it better
depends on having some gold standard

Peter

Peter L. Flom, PhD
Statistical Consultant
Website: www DOT peterflomconsulting DOT com
Writing; http://www.associatedcontent.com/user/582880/peter_flom.html
Twitter: @peterflom

[Heart]

The grading describe the amount of a pathological situation that
represents a risk factor for both TIA and stroke.
Higher the grade, higher the disease severity (higher the grade,
higher the stroke/TIA ratio).
There are two grading scale that help us to quantitatively evaluate
this risk factor, but one, in my opinion, is more acceptable because
the stroke/TIA ratios distribute more proportionally among the grades
(similar to a linear regression).
My trouble is translate this observation into a statistical test.



On 6 Ott, 16:42, Peter Flom <peterflomconsult...@mindspring.com>
wrote:
> Heart <nicmarch...@gmail.com> wrote