wuzzy <[EMAIL PROTECTED]> wrote:
> It is clear that sampling a person for a serum value in one shot will
> not give you a correct account of the population's serum distribution 
> My own belief is that you *can* use single-shot serum values to
> predict disease risk:  but you have to use group these with a value
> that has areliable distribution: ex. group serum cholesterol by
> subject's bodyweight, which does not vary appreciably among
> individuals.

> For instance divide your subjects into those with high bodyweight and
> low bodyweight and use t-test to compare means of cholesterol levels. 
> Is there any criticisms to the above: the t-test should be sufficient
> Any info, references (journals or books) appreciated.  

You will still have dilution of the true strength of the association.
See Qizilbash et al AJE 1991 133:832-8 for one example of a case-control
study with modelling of measurement error.  Bollen _Structural equations
with latent variables_ is a nice intro to measurement models.  There are
a lot of Bayesian approaches to the problem.  A Medline search for
"measurement error relative risk" will give some recent examples eg
Rosner & Gore Am J Epidemiol 2001 154:827-35 (Rosner has written much on
the topic).

You can use a single measurement to make disease risk assessment, --
there are lots of papers on this eg single random blood glucose criteria
for diabetes, single blood pressure reading for hypertension, single
cholesterol measurement for CHD risk.


-- 
| David Duffy.                                                     ,-_|\
| email: [EMAIL PROTECTED]  ph: INT+61+7+3362-0217 fax: -0101    /     *
| Epidemiology Unit, The Queensland Institute of Medical Research \_,-._/
| 300 Herston Rd, Brisbane, Queensland 4029, Australia                 v 
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to