I'll offer a response David Atkin's question about the use of LRs in
different populations.
In short: I believe it is important to recognize WHY the population is
different. That is, if the population is different because they have
been screened by a physician first, then the LRs MAY be invalid. If
they have not been screened, the LRs should be robust.
To explain further: Consider a 2X2 table. LRs will remain robust if the
population's distribution is changed only from one column to another.
This has been assumed to be true for populations that have not been
"filtered" by contact with a physician.
However, physicians filter and refer patients according to their
hypothesis testing with data from the history, physical exam, and
initial lab testing. This implies that the population referred on for
further evaluation will be enriched with patients from the positive row,
most heavily from the true positive cell. In this case, sensitivity
will rise and specificity will fall. This will obviously effect the
LRs, usually causing a drop.
An important assumption that needs to be tested is whether prevalence
is, in fact, the only important parameter that changes in unfiltered
populations. I am unaware of any studies of this.
We have been looking at a metanalysis of the history and physical exams
of patients being evaluated for appendicitis at various points along the
"referral continuum". Interestingly, it appears LRs may also RISE in
referred populations. We speculate this is because the initial
physician is not as well trained in the the physical exam technique. It
is too early in our analysis to share specifics, but we hope to have
more definitive data to share soon.
Jim
James M. Wagner, MD
Assoc. Professor, Internal Medicine
U. Texas Southwestern Medical School
Dallas, TX, USA
214 648 2168
[log in to unmask]
>>> Dave Sackett <[log in to unmask]> 09/22 9:43 AM >>>
you're best qualified to answer this
cheers
dls
............................................................................
Prof David L. Sackett
Director, NHS R&D Centre for Evidence-Based Medicine
Consultant in Medicine Editor, Evidence-Based
Medicine
Nuffield Department of Medicine, University of Oxford
Level 5, John Radcliffe Hospital, Oxford OX3 9DU, England
Phone: +44-(0)1865-221320 Fax: +44-(0)1865 222901
Email: [log in to unmask] WWW: http://cebm.jr2.ox.ac.uk
............................................................................
---------- Forwarded message ----------
Date: Fri, 18 Sep 1998 11:21:06 -0400
From: "Atkins, David" <[log in to unmask]>
To: Evidence Based Health Group <[log in to unmask]>
Subject: likelihood ratios
In light of the discussion a few weeks ago about the stability of
sensitivity and specificity in different populations, and a very nice
review of diagnostic test parameters by DLS in acp journal club this
week, I have questions about the use of likelihood ratios and nomograms
of the sort cited from Fagan. If sensitivity and specificity of a given
test varies in different populations (where spectrum of disease and
confounding conditions will vary), couldn't it be misleading to take
likelihood ratios derived from a broad population (e.g. ferritin levels
from a consecutive series of patients evaluated for anemia in your
clinic) and apply them to a specific patient for whom one has estimated
pretest probability of anemia based on clinical judgment (e.g.
menstruating young female complaining of fatigue)? Don't we need to
know sensitivity and specificity of a given ferritin level in a
population of menstruating women with fatigue, at which point we can
calculate post-test probability without relying on estimates of pretest
probability and likelihood ratios? It seems that likelihood ratios rely
on the assumption that sensitivity and specificity are constant across
distint spectrums of patients and disease severity, which last weeks
discussion implied was demonstrably not the case. Any thoughts?
(Please note new street address and phone #)
David Atkins, MD, MPH
Center for Practice and Technology Assessment
Agency for Health Care Policy and Research
6010 Executive Blvd, Suite 300
Rockville, MD
(301) 594-4016
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|